backend: let a model ask for a line break, and read the 参考文献 list back

The paper model gained two things while the MCP layer was not looking, and
both were invisible from a tool call.

`paper_sentence.line_break_before` — the flag a paragraph needs so a formula,
a list of items or an author line can start a line of its own — was not in
`SentenceSpec`, and `sentence_inputs` dropped it on the floor: a model could
write the field, watch the call succeed, and get prose back. It is now part of
every write path (`paragraph_write`, `paper_write`, `sentence_add`,
`sentence_update`) and of every read path (`sentence_row` reports it, so a
model can see the lines before moving one). Two bugs fell out of doing it
properly:

* `paragraph_move` rebuilt its sentences without the flag, so moving a
  paragraph that carried a formula turned it into prose. It also passed the
  *source* sorts through, which interleaved the moved sentences with the
  target's own — the merged paragraph read in an order neither writer wrote.
  Both are fixed; the tool now appends, as its own description promised.
* `document_text` printed a formula run into the sentence before it. The paper
  view renders `$$…$$` as a block, so the seam rules are now the same in both
  places, with a comment on each side naming the other.

`PaperDocumentRead.references` and `reference_style` — the per-work 参考文献
list the backend now assembles and formats — were reachable over REST but not
over MCP: `paper_document(format="json")` returned citations and nothing to
point them at, even though docs/MCP.md already described the list. The json
form now carries the entries (index, reference_id, text, missing, citations)
and the style name, so a model quotes the bibliography instead of assembling a
second one out of the same fields.

smoke_mcp.py grows from 76 checks to 99 over two new steps — the line-break
step pins the display-formula rule, the marker toggle and the moving case; the
bibliography step cites one work from two sentences and checks that the text,
the entry and both markers agree on one number.
This commit is contained in:
2026-09-23 11:49:39 +08:00
parent 960450f744
commit 9f2bbe50eb
11 changed files with 483 additions and 34 deletions
+19
View File
@@ -45,6 +45,16 @@ class SentenceSpec(BaseModel):
"""One sentence, optionally carrying citations.""" """One sentence, optionally carrying citations."""
content: Annotated[str, Field(description="句子正文;空白会被折叠成一行")] content: Annotated[str, Field(description="句子正文;空白会被折叠成一行")]
line_break_before: Annotated[
bool,
Field(
description=(
"这一句前面另起一行(默认 false)。正文被句子拼成一段,所以公式、"
"分条列举、作者行这类不是散文的行要在这里开口;break 会顶掉句间的空格,"
"不会多印一个空格。display 公式($$…$$ 或 \\[…\\])本来就独占一行,不必设这个"
)
),
] = False
citations: Annotated[ citations: Annotated[
list[CitationSpec], list[CitationSpec],
Field(description="这一段引用了哪些文献"), Field(description="这一段引用了哪些文献"),
@@ -59,6 +69,15 @@ class TextSpec(BaseModel):
SplitMode, SplitMode,
Field(description="切句方式:line 一行一句(默认)/ sentence 按句号切 / paragraph 整段一句"), Field(description="切句方式:line 一行一句(默认)/ sentence 按句号切 / paragraph 整段一句"),
] = "line" ] = "line"
line_break_before: Annotated[
bool,
Field(
description=(
"这一整块从新的一行开始(默认 false);只作用于切出来的第一句,"
"因为切出来的其余几句是接着它的,不是各自另起一行"
)
),
] = False
citations: Annotated[ citations: Annotated[
list[CitationSpec], list[CitationSpec],
Field(description="整段共用的引用"), Field(description="整段共用的引用"),
+59 -4
View File
@@ -256,6 +256,10 @@ def sentence_row(sentence: Any, *, separator: str | None = None) -> dict:
"position": sentence.paper_template_filed_sort, "position": sentence.paper_template_filed_sort,
"sort": sentence.sort, "sort": sentence.sort,
"content": sentence.content, "content": sentence.content,
# Reported so a writer can see where the lines fall before deciding to
# move one: a paragraph read back without this flag looks like prose
# that happens to have no spaces in it.
"line_break_before": sentence.line_break_before,
"citations": [citation_row(item) for item in sentence.citations], "citations": [citation_row(item) for item in sentence.citations],
} }
if separator is not None: if separator is not None:
@@ -277,6 +281,21 @@ def paragraph_row(paragraph: ParagraphRead, *, preview: int = 80) -> dict:
} }
#: A formula the paper view sets on a line of its own: ``$$…$$`` or ``\[…\]``.
#:
#: This is the same rule as ``hasDisplayMath`` in ``frontend/src/utils/math.ts``,
#: written twice because the view is TypeScript and this is Python. The two must
#: agree, or a tool result and the page a writer is looking at would disagree
#: about where the lines are — which is the one thing this module exists to
#: prevent. Change one, change the other.
_DISPLAY_MATH = re.compile(r"\$\$[\s\S]+?\$\$|\\\[[\s\S]+?\\\]")
def _is_display(sentence: Any) -> bool:
"""Whether this sentence is a display formula — a block, not a run of text."""
return bool(_DISPLAY_MATH.search(sentence.content))
def paragraph_text( def paragraph_text(
paragraph: ParagraphRead, paragraph: ParagraphRead,
*, *,
@@ -289,6 +308,14 @@ def paragraph_text(
belongs at each seam, so a tool result and the paper view agree on where belongs at each seam, so a tool result and the paper view agree on where
the spaces go. Nothing here adds spacing of its own. the spaces go. Nothing here adds spacing of its own.
A line break is printed where the model asked for one, and around a display
formula — the paper view renders ``$$…$$`` as a block, so the lines on
either side of it are line boundaries there too. A break *replaces* the
separator rather than joining it: the new line already separates the two
sentences, and a space as well would indent the line below by one blank.
The first sentence of a paragraph is never preceded by a break, because
there is nothing above it to break from.
A marker's number comes from ``index_by_citation`` — the citation's *work* A marker's number comes from ``index_by_citation`` — the citation's *work*
number as the whole document assigned it, not its position in this sentence — number as the whole document assigned it, not its position in this sentence —
so what is printed in the text matches the 参考文献 list it points at. Read so what is printed in the text matches the 参考文献 list it points at. Read
@@ -296,8 +323,15 @@ def paragraph_text(
falls back to the citation's place in the sentence. falls back to the citation's place in the sentence.
""" """
parts: list[str] = [] parts: list[str] = []
previous_block = False
for sentence in paragraph.sentences: for sentence in paragraph.sentences:
parts.append(sentence.separator_before + sentence.content) block = _is_display(sentence)
if parts and (sentence.line_break_before or block or previous_block):
parts.append("\n")
else:
parts.append(sentence.separator_before)
parts.append(sentence.content)
previous_block = block
if citations and sentence.citations: if citations and sentence.citations:
numbers = [ numbers = [
str( str(
@@ -592,6 +626,8 @@ def sentence_inputs(items: Sequence[Any], *, default_split: str = "line") -> lis
* ``"A sentence."`` — the common case; * ``"A sentence."`` — the common case;
* ``{"content": "...", "citations": [{"quote": "...", "reference_id": 3}]}``; * ``{"content": "...", "citations": [{"quote": "...", "reference_id": 3}]}``;
* ``{"content": "$$ y = ax + b $$", "line_break_before": true}`` — a block
that starts a line of its own;
* ``{"text": "one\\ntwo", "split": "line"}`` — a block to be cut. * ``{"text": "one\\ntwo", "split": "line"}`` — a block to be cut.
Every path ends in :class:`~app.schemas.paper.SentenceInput`, so the Every path ends in :class:`~app.schemas.paper.SentenceInput`, so the
@@ -621,15 +657,34 @@ def sentence_inputs(items: Sequence[Any], *, default_split: str = "line") -> lis
) )
for citation in (item.get("citations") or []) for citation in (item.get("citations") or [])
] ]
line_break = bool(item.get("line_break_before", False))
if item.get("content") is not None: if item.get("content") is not None:
out.append(SentenceInput(content=str(item["content"]), citations=citations)) out.append(
SentenceInput(
content=str(item["content"]),
line_break_before=line_break,
citations=citations,
)
)
continue continue
if item.get("text") is not None: if item.get("text") is not None:
mode = item.get("split") or default_split mode = item.get("split") or default_split
for piece in split_text(str(item["text"]), mode): pieces = split_text(str(item["text"]), mode)
out.append(SentenceInput(content=piece, citations=citations)) for offset, piece in enumerate(pieces):
# A break asked for on a *block* belongs to its first sentence:
# "this block starts a new line". The pieces after it are
# separated by whatever the split found, and flagging them too
# would turn a paragraph the writer wrapped across three lines
# into three lines of the paper.
out.append(
SentenceInput(
content=piece,
line_break_before=line_break and offset == 0,
citations=citations,
)
)
continue continue
fail("句子对象需要有 content 或 text 字段") fail("句子对象需要有 content 或 text 字段")
+24 -2
View File
@@ -248,7 +248,7 @@ def register(server: MCPServer) -> None:
bool, Field(description="是否带上引用标记与参考文献列表") bool, Field(description="是否带上引用标记与参考文献列表")
] = True, ] = True,
) -> str: ) -> str:
"""读整篇论文:按模板顺序输出全文(含引用编号和参考文献)。""" """读整篇论文:按模板顺序输出全文(含引用编号、参考文献表与换行)。"""
with s.session() as db: with s.session() as db:
paper = s.paper_or_fail(db, paper_id) paper = s.paper_or_fail(db, paper_id)
document = paper_crud.build_document(db, paper) document = paper_crud.build_document(db, paper)
@@ -270,6 +270,8 @@ def register(server: MCPServer) -> None:
for paragraph in document.paragraphs for paragraph in document.paragraphs
if include_empty or paragraph.sentences if include_empty or paragraph.sentences
] ]
# One row per *place* in the text, each carrying the number of
# the work it cites — the number the marker in the text shows.
payload["citations"] = [ payload["citations"] = [
{ {
"index": item.index, "index": item.index,
@@ -279,6 +281,24 @@ def register(server: MCPServer) -> None:
} }
for item in document.citations for item in document.citations
] ]
# The 参考文献 list as it prints: one entry per work, already
# formatted by the server, so a model quotes the bibliography
# rather than assembling a second one out of the same fields.
# It is reported even when it is empty, because "this paper
# cites nothing yet" is an answer.
payload["references"] = [
{
"index": entry.index,
"reference_id": entry.reference_id,
"text": entry.text,
"missing": entry.missing,
# The ids of the places that cite it; the rows
# themselves are in citations above.
"citations": [row.id for row in entry.citations],
}
for entry in document.references
]
payload["reference_style"] = document.reference_style
if document.warnings: if document.warnings:
payload["warnings"] = document.warnings payload["warnings"] = document.warnings
return s.dumps(payload) return s.dumps(payload)
@@ -291,7 +311,8 @@ def register(server: MCPServer) -> None:
Field( Field(
description=( description=(
"要写的段落数组。每段用 heading(如 “1. Introduction”)或 position 指定," "要写的段落数组。每段用 heading(如 “1. Introduction”)或 position 指定,"
"正文放 sentences(字符串数组或带 citations 的对象数组)," "正文放 sentences(字符串数组或带 citations 的对象数组,"
"要另起一行的公式或条目加 line_break_before: true),"
"或放 text + split 让服务端切句" "或放 text + split 让服务端切句"
) )
), ),
@@ -324,6 +345,7 @@ def register(server: MCPServer) -> None:
SentenceCreate( SentenceCreate(
paper_template_filed_sort=position, paper_template_filed_sort=position,
content=item.content, content=item.content,
line_break_before=item.line_break_before,
citations=item.citations, citations=item.citations,
sort=item.sort, sort=item.sort,
), ),
+21 -3
View File
@@ -74,7 +74,7 @@ def register(server: MCPServer) -> None:
str | None, Field(description="段落标题,如 “2. Methods”") str | None, Field(description="段落标题,如 “2. Methods”")
] = None, ] = None,
) -> str: ) -> str:
"""读一个段落:拼好的正文、逐句内容与 id、引用、段间分隔符。""" """读一个段落:拼好的正文、逐句内容与 id、引用、段间分隔符与换行。"""
with s.session() as db: with s.session() as db:
paper = s.paper_or_fail(db, paper_id) paper = s.paper_or_fail(db, paper_id)
field_sort = s.resolve_position(paper, position=position, heading=heading) field_sort = s.resolve_position(paper, position=position, heading=heading)
@@ -109,7 +109,13 @@ def register(server: MCPServer) -> None:
] = None, ] = None,
sentences: Annotated[ sentences: Annotated[
list[SentenceItem] | None, list[SentenceItem] | None,
Field(description="句子数组:字符串、{content, citations} 或 {text, split}"), Field(
description=(
"句子数组:字符串、{content, citations} 或 {text, split}。"
"要某一句另起一行(公式、分条列举、作者行),用对象形式加上 "
"line_break_before: true"
)
),
] = None, ] = None,
text: Annotated[ text: Annotated[
str | None, Field(description="整段文本;与 sentences 二选一") str | None, Field(description="整段文本;与 sentences 二选一")
@@ -199,7 +205,18 @@ def register(server: MCPServer) -> None:
s.fail(f"第 {source} 段没有内容可以搬") s.fail(f"第 {source} 段没有内容可以搬")
inputs = [ inputs = [
SentenceInput(content=row.content, sort=row.sort, citations=list(row.citations)) SentenceInput(
content=row.content,
# ``sort`` is deliberately left unset so that
# :func:`app.crud.paper.replace_paragraph` numbers the block
# from the *target's* current maximum. Carrying the old
# numbers over instead would interleave the two paragraphs —
# the moved sentence with ``sort=1`` would sort against the
# target's own sentence with ``sort=1`` — and the merge would
# read in an order neither writer wrote.
line_break_before=row.line_break_before,
citations=list(row.citations),
)
for row in rows for row in rows
] ]
paper_crud.replace_paragraph( paper_crud.replace_paragraph(
@@ -258,6 +275,7 @@ def _write(
SentenceCreate( SentenceCreate(
paper_template_filed_sort=field_sort, paper_template_filed_sort=field_sort,
content=item.content, content=item.content,
line_break_before=item.line_break_before,
citations=item.citations, citations=item.citations,
sort=item.sort, sort=item.sort,
), ),
+23 -2
View File
@@ -85,6 +85,15 @@ def register(server: MCPServer) -> None:
citations: Annotated[ citations: Annotated[
list[CitationSpec] | None, Field(description="这句引用了哪些文献") list[CitationSpec] | None, Field(description="这句引用了哪些文献")
] = None, ] = None,
line_break_before: Annotated[
bool,
Field(
description=(
"这一句前面另起一行(默认 false):公式、分条列举、作者行用得上。"
"break 会顶掉句间的空格"
)
),
] = False,
sort: Annotated[ sort: Annotated[
int | None, Field(description="段内位置;不传则排在段尾") int | None, Field(description="段内位置;不传则排在段尾")
] = None, ] = None,
@@ -99,6 +108,7 @@ def register(server: MCPServer) -> None:
SentenceCreate( SentenceCreate(
paper_template_filed_sort=field_sort, paper_template_filed_sort=field_sort,
content=content, content=content,
line_break_before=line_break_before,
citations=_citations(citations or []), citations=_citations(citations or []),
sort=sort, sort=sort,
), ),
@@ -123,13 +133,23 @@ def register(server: MCPServer) -> None:
sort: Annotated[ sort: Annotated[
int | None, Field(description="段内排序值;不传则不动") int | None, Field(description="段内排序值;不传则不动")
] = None, ] = None,
line_break_before: Annotated[
bool | None,
Field(description="这一句是否另起一行;不传则不动,false 表示取消换行"),
] = None,
citations: Annotated[ citations: Annotated[
list[CitationSpec] | None, list[CitationSpec] | None,
Field(description="引用整组替换;传 [] 表示清空引用"), Field(description="引用整组替换;传 [] 表示清空引用"),
] = None, ] = None,
) -> str: ) -> str:
"""修改一句话的正文、所属段落、排序或引用(引用是整体替换)。""" """修改一句话的正文、所属段落、排序、换行或引用(引用是整体替换)。"""
if content is None and position is None and sort is None and citations is None: if (
content is None
and position is None
and sort is None
and line_break_before is None
and citations is None
):
s.fail("没有给出任何要修改的内容") s.fail("没有给出任何要修改的内容")
with s.session() as db: with s.session() as db:
@@ -143,6 +163,7 @@ def register(server: MCPServer) -> None:
content=content, content=content,
paper_template_filed_sort=position, paper_template_filed_sort=position,
sort=sort, sort=sort,
line_break_before=line_break_before,
citations=None if citations is None else _citations(citations), citations=None if citations is None else _citations(citations),
), ),
) )
+274 -7
View File
@@ -152,7 +152,9 @@ async def run(url: str | None = None, token: str | None = None) -> int:
async def _run(url: str | None = None, token: str | None = None) -> int: async def _run(url: str | None = None, token: str | None = None) -> int:
template_id, outline = pick_template() template_id, outline = pick_template()
check("a template with fields exists", len(outline) >= 3, f"got {len(outline)}") # Four headings, because the line-break and bibliography steps below move one
# paragraph onto a fourth; with fewer, they would have nowhere to write.
check("a template with four fields exists", len(outline) >= 4, f"got {len(outline)}")
first = outline[0]["name"] first = outline[0]["name"]
second = outline[1]["name"] second = outline[1]["name"]
@@ -592,12 +594,277 @@ async def _run(url: str | None = None, token: str | None = None) -> int:
for row in paragraph["sentences"] for row in paragraph["sentences"]
), ),
) )
check(
"it names the citation style the entries are printed in",
structured["reference_style"] == "springer",
str(structured.get("reference_style")),
)
# Step 13 rewrote every paragraph, so the citation step 9 wrote is gone
# and this paper cites nothing: the 参考文献 list is reported as empty
# rather than left out, because "this cites nothing yet" is an answer.
check(
"and reports the 参考文献 list as empty when nothing is cited",
structured["references"] == [],
json.dumps(structured["references"], ensure_ascii=False),
)
print("\n15. search across papers") print("\n15. search across papers")
found = await call(session, "sentence_search", keyword="正文写在标题下面", paper_id=paper["id"]) found = await call(session, "sentence_search", keyword="正文写在标题下面", paper_id=paper["id"])
check("the search finds what was written", found["total"] >= 1, str(found["total"])) check("the search finds what was written", found["total"] >= 1, str(found["total"]))
print("\n16. move a paragraph, then clear the one it moved onto") print("\n16. the writer asks for a line break")
# A paragraph is its sentences run together, so a formula, a list of
# items or an author line has to say where the lines are. A *display*
# formula does not have to: the paper view sets one on a line of its own
# by definition, and the text a tool hands back has to agree with the
# page the writer is looking at, or the two disagree about the lines.
prose = outline[2]
await call(
session,
"paper_write",
paper_id=paper["id"],
paragraphs=[
{
"heading": prose["name"],
"sentences": [
"本段说明公式的来历。",
{"content": "$$ E = mc^2 $$"},
{"content": "其中 c 为光速。", "line_break_before": True},
],
}
],
)
written = await call(
session, "paragraph_get", paper_id=paper["id"], heading=prose["name"]
)
check(
"the flag is stored per sentence",
[row["line_break_before"] for row in written["sentences"]]
== [False, False, True],
json.dumps(written["sentences"], ensure_ascii=False),
)
check(
"a formula takes its own line unasked, and a flagged sentence too",
written["text"] == "本段说明公式的来历。\n$$ E = mc^2 $$\n其中 c 为光速。",
repr(written["text"]),
)
check(
"and the flag replaces the separator rather than joining it",
"\n " not in written["text"],
repr(written["text"]),
)
whole = await call(session, "paper_document", paper_id=paper["id"])
check(
"the whole document prints it the same way",
"本段说明公式的来历。\n$$ E = mc^2 $$\n其中 c 为光速。" in whole["text"],
repr(whole["text"][-260:]),
)
added = await call(
session,
"sentence_add",
paper_id=paper["id"],
heading=prose["name"],
content="另一条列举。",
line_break_before=True,
)
check(
"sentence_add can ask for one as well",
added["sentence"]["line_break_before"] is True,
json.dumps(added["sentence"], ensure_ascii=False),
)
# Toggled on the sentence that follows the prose, not the one that
# follows the formula: the formula is a block, so the sentence after it
# starts a line whatever its flag says, and a check written on that one
# would pass for the wrong reason.
joined = added["sentence"]["id"]
off = await call(
session,
"sentence_update",
paper_id=paper["id"],
sentence_id=joined,
line_break_before=False,
)
check(
"the break can be taken off again",
off["sentence"]["line_break_before"] is False,
json.dumps(off, ensure_ascii=False),
)
rejoined = await call(
session, "paragraph_get", paper_id=paper["id"], heading=prose["name"]
)
check(
"and the two sentences run on as one line",
"\n另一条列举。" not in rejoined["text"]
and "其中 c 为光速。另一条列举。" in rejoined["text"],
repr(rejoined["text"]),
)
check(
"while the line after the formula stays broken, flag or no flag",
"\n其中 c 为光速。" in rejoined["text"],
repr(rejoined["text"]),
)
await call(
session,
"sentence_update",
paper_id=paper["id"],
sentence_id=joined,
line_break_before=True,
)
relined = await call(
session, "paragraph_get", paper_id=paper["id"], heading=prose["name"]
)
check(
"asking for it again puts the line back",
"\n另一条列举。" in relined["text"],
repr(relined["text"]),
)
moved_texts = [row["content"] for row in written["sentences"]] + [
added["sentence"]["content"]
]
await call(
session,
"paragraph_move",
paper_id=paper["id"],
heading=prose["name"],
target_heading=outline[3]["name"],
)
landed = await call(
session, "paragraph_get", paper_id=paper["id"], heading=outline[3]["name"]
)
# Matched by content rather than by id: a move rewrites the rows, so the
# ids the source paragraph had do not survive it. Citations are rebuilt
# the same way, which is why they come back attached to the new rows.
flag_of = {row["content"]: row["line_break_before"] for row in landed["sentences"]}
check(
"a paragraph carries its line breaks when it moves",
[flag_of.get(text) for text in moved_texts] == [False, False, True, True],
f"moved={moved_texts} reads back as "
f"{[flag_of.get(text) for text in moved_texts]} out of "
f"{json.dumps(landed['sentences'], ensure_ascii=False)}",
)
check(
"and it lands after the target's own sentences, as the tool says",
[row["content"] for row in landed["sentences"]][-len(moved_texts):]
== moved_texts,
json.dumps([row["content"] for row in landed["sentences"]], ensure_ascii=False),
)
print("\n17. the 参考文献 list is printed, not assembled by the client")
# Step 10 deleted the reference this run made, so the bibliography step
# brings its own work to cite. It is cited twice on purpose: one work,
# one entry, one number in both places is the rule being checked.
bibliography = await call(
session,
"reference_create",
reference={
"type": "journal-article",
"title": "SMOKE MCP 参考文献表",
"container_title": "Journal of Smoke Tests",
"publication_year": 2024,
"volume": "12",
"issue": "3",
"pages": "45-67",
"doi": "10.1000/smoke.mcp.002",
"authors": [
{"name": "Smith, John", "family_name": "Smith", "given_name": "John"}
],
},
)
cited_reference = bibliography["id"]
_created_references.append(cited_reference)
first = await call(
session,
"sentence_add",
paper_id=paper["id"],
heading=outline[3]["name"],
content="A cited sentence.",
citations=[{"quote": "引用的原话", "reference_id": cited_reference}],
)
await call(
session,
"sentence_add",
paper_id=paper["id"],
heading=outline[3]["name"],
content="A second sentence, same work.",
line_break_before=True,
citations=[{"quote": "再次引用", "reference_id": cited_reference}],
)
document = await call(
session, "paper_document", paper_id=paper["id"], format="json"
)
check(
"the style the entries are written in is named",
document["reference_style"] == "springer",
str(document.get("reference_style")),
)
check(
"one row per place in the text",
len(document["citations"]) == 2,
json.dumps(document["citations"], ensure_ascii=False),
)
check(
"but one entry for the work, however often it is cited",
len(document["references"]) == 1,
json.dumps(document["references"], ensure_ascii=False),
)
entry = document["references"][0]
check(
"the entry is a bibliography line rather than the quoted words",
entry["text"].startswith("Smith, J.: SMOKE MCP 参考文献表")
and "Journal of Smoke Tests 12(3), 45–67 (2024)" in entry["text"]
and entry["text"].endswith("https://doi.org/10.1000/smoke.mcp.002"),
entry["text"],
)
check(
"a complete entry is reported as complete",
entry["missing"] == [],
str(entry["missing"]),
)
check(
"both places wear the one number the entry has",
{row["index"] for row in document["citations"]} == {entry["index"]},
json.dumps(document["citations"], ensure_ascii=False),
)
check(
"and the entry names both of them",
len(entry["citations"]) == 2
and first["sentence"]["id"]
in {row["sentence_id"] for row in document["citations"]},
json.dumps(entry, ensure_ascii=False),
)
as_text = await call(session, "paper_document", paper_id=paper["id"])
check(
"the text form carries the 参考文献 block with that entry",
f"参考文献\n[{entry['index']}] {entry['text']}" in as_text["text"],
repr(as_text["text"][-300:]),
)
check(
"and marks both places with that same number",
f"A cited sentence.[{entry['index']}]" in as_text["text"]
and f"A second sentence, same work.[{entry['index']}]" in as_text["text"],
repr(as_text["text"][-700:]),
)
check(
"and the bibliography lists the work exactly once",
len(
[
line
for line in as_text["text"].splitlines()
if line.startswith(f"[{entry['index']}] ")
]
)
== 1,
repr(as_text["text"][-300:]),
)
print("\n18. move a paragraph, then clear the one it moved onto")
home = outline[0] home = outline[0]
target = outline[1] target = outline[1]
relocation = await call( relocation = await call(
@@ -629,7 +896,7 @@ async def _run(url: str | None = None, token: str | None = None) -> int:
after_clear = await call(session, "paragraph_get", paper_id=paper["id"], position=target["position"]) after_clear = await call(session, "paragraph_get", paper_id=paper["id"], position=target["position"])
check("and the paragraph is empty", after_clear["sentences"] == [], json.dumps(after_clear, ensure_ascii=False)) check("and the paragraph is empty", after_clear["sentences"] == [], json.dumps(after_clear, ensure_ascii=False))
print("\n17. the field library round-trips") print("\n19. the field library round-trips")
field = await call(session, "field_create", name="SMOKE MCP 字段", level=2, font_size=9.5) field = await call(session, "field_create", name="SMOKE MCP 字段", level=2, font_size=9.5)
check("created with its typography", field["level"] == 2 and field["font_size"] == 9.5, json.dumps(field, ensure_ascii=False)) check("created with its typography", field["level"] == 2 and field["font_size"] == 9.5, json.dumps(field, ensure_ascii=False))
renamed = await call(session, "field_update", field_id=field["id"], name="SMOKE MCP 字段(改)") renamed = await call(session, "field_update", field_id=field["id"], name="SMOKE MCP 字段(改)")
@@ -638,7 +905,7 @@ async def _run(url: str | None = None, token: str | None = None) -> int:
check("deleted while no template places it", dropped["deleted"] == 1, json.dumps(dropped, ensure_ascii=False)) check("deleted while no template places it", dropped["deleted"] == 1, json.dumps(dropped, ensure_ascii=False))
check("and it is reported as found, not missing", dropped["missing"] == [], json.dumps(dropped, ensure_ascii=False)) check("and it is reported as found, not missing", dropped["missing"] == [], json.dumps(dropped, ensure_ascii=False))
print("\n18. switch the template and keep the content") print("\n20. switch the template and keep the content")
spare = await call( spare = await call(
session, session,
"template_create", "template_create",
@@ -682,15 +949,15 @@ async def _run(url: str | None = None, token: str | None = None) -> int:
json.dumps(switched_document, ensure_ascii=False)[:400], json.dumps(switched_document, ensure_ascii=False)[:400],
) )
print("\n19. a template in use cannot be deleted") print("\n21. a template in use cannot be deleted")
message = await expect_error(session, "template_delete", template_ids=spare["id"]) message = await expect_error(session, "template_delete", template_ids=spare["id"])
check("the refusal names the blocker", "SMOKE MCP 模板" in message, message) check("the refusal names the blocker", "SMOKE MCP 模板" in message, message)
print("\n20. deleting a paper needs an explicit confirmation") print("\n22. deleting a paper needs an explicit confirmation")
message = await expect_error(session, "paper_delete", paper_ids=paper["id"]) message = await expect_error(session, "paper_delete", paper_ids=paper["id"])
check("the refusal explains itself", "confirm=true" in message, message) check("the refusal explains itself", "confirm=true" in message, message)
print("\n21. delete the paper and the template") print("\n23. delete the paper and the template")
deleted = await call(session, "paper_delete", paper_ids=paper["id"], confirm=True) deleted = await call(session, "paper_delete", paper_ids=paper["id"], confirm=True)
check("deleted", deleted["deleted"] == 1, json.dumps(deleted, ensure_ascii=False)) check("deleted", deleted["deleted"] == 1, json.dumps(deleted, ensure_ascii=False))
_created_papers.remove(paper["id"]) _created_papers.remove(paper["id"])
+5 -1
View File
@@ -18,7 +18,11 @@ paper-doc 的 MCP 服务跑在 k3s 里,命名空间 **`paper-doc`**,客户
| Secret | `paper-doc-secrets` | `DB_USER` / `DB_PASSWORD` / `MCP_HTTP_TOKEN` | | Secret | `paper-doc-secrets` | `DB_USER` / `DB_PASSWORD` / `MCP_HTTP_TOKEN` |
| Secret | `aliyun-acr` | 镜像仓库拉取凭证(从 `default` 复制) | | Secret | `aliyun-acr` | 镜像仓库拉取凭证(从 `default` 复制) |
镜像:`registry.cn-hangzhou.aliyuncs.com/mcp_skill/paper-doc-mcp:v2` 镜像:`registry.cn-hangzhou.aliyuncs.com/mcp_skill/paper-doc-mcp:v5`
(`sha256:1e17a72ca611a498d93fc960a7fe1eba5ed5803376794efa66f1967830d51c5d`;
`kubectl -n paper-doc get pods -o jsonpath='{.items[0].status.containerStatuses[0].imageID}'`
对得上,才说明跑的就是 ACR 上那一版。历史:v1 首发布 → v2 重命名表 → v3 参考文献库 →
v4 补参考文献 MCP 工具 → **v5 换行与参考文献表**。)
**数据库**:TiDB 已经在这个集群里,Deployment 用集群内的 Service **数据库**:TiDB 已经在这个集群里,Deployment 用集群内的 Service
`tidb-tidb.tidb:4000`,而不是宿主机用的 NodePort `192.168.1.88:32738`。 `tidb-tidb.tidb:4000`,而不是宿主机用的 NodePort `192.168.1.88:32738`。
+4 -2
View File
@@ -128,6 +128,8 @@ log "done: $IMAGE"
cat <<EOF cat <<EOF
Next: Next:
kubectl -n paper_doc set image deployment/paper-doc-mcp mcp=$IMAGE kubectl -n paper-doc apply -f deploy/k8s/10-paper-doc-mcp.yaml # 或 set image,tag 要先改进去
kubectl -n paper_doc rollout status deployment/paper-doc-mcp kubectl -n paper-doc set image deployment/paper-doc-mcp mcp=$IMAGE
kubectl -n paper-doc rollout status deployment/paper-doc-mcp
# 换镜像后 DSH 的 mcp-client 插件会攥着旧 Pod 的会话,记得触发热加载重连
EOF EOF
+3 -3
View File
@@ -67,7 +67,7 @@ spec:
type: RuntimeDefault type: RuntimeDefault
containers: containers:
- name: mcp - name: mcp
image: registry.cn-hangzhou.aliyuncs.com/mcp_skill/paper-doc-mcp:v4 image: registry.cn-hangzhou.aliyuncs.com/mcp_skill/paper-doc-mcp:v5
imagePullPolicy: IfNotPresent imagePullPolicy: IfNotPresent
ports: ports:
- name: http - name: http
@@ -178,13 +178,13 @@ spec:
# imagePullSecrets: [{name: aliyun-acr}] # imagePullSecrets: [{name: aliyun-acr}]
# initContainers: # initContainers:
# - name: migrate # - name: migrate
# image: registry.cn-hangzhou.aliyuncs.com/mcp_skill/paper-doc-mcp:v4 # image: registry.cn-hangzhou.aliyuncs.com/mcp_skill/paper-doc-mcp:v5
# command: ["alembic", "upgrade", "head"] # command: ["alembic", "upgrade", "head"]
# envFrom: [{configMapRef: {name: paper-doc-config}}] # envFrom: [{configMapRef: {name: paper-doc-config}}]
# env: [...] # the same three secret keys as above # env: [...] # the same three secret keys as above
# containers: # containers:
# - name: api # - name: api
# image: registry.cn-hangzhou.aliyuncs.com/mcp_skill/paper-doc-mcp:v4 # image: registry.cn-hangzhou.aliyuncs.com/mcp_skill/paper-doc-mcp:v5
# command: ["uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "8000"] # command: ["uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "8000"]
# ports: [{name: http, containerPort: 8000}] # ports: [{name: http, containerPort: 8000}]
# envFrom: [{configMapRef: {name: paper-doc-config}}] # envFrom: [{configMapRef: {name: paper-doc-config}}]
+49 -9
View File
@@ -1,13 +1,13 @@
# MCP 服务(paper-doc-mcp) # MCP 服务(paper-doc-mcp)
把 paper-doc 的论文、段落、句子、模板、字段库能力暴露成 **Model Context Protocol** 工具, 把 paper-doc 的论文、段落、句子、模板、字段库、文献库能力暴露成 **Model Context Protocol** 工具,
让 Claude Code、Codex、DeepSeek Harness(DSH)等任何 MCP 客户端直接增删改查论文。 让 Claude Code、Codex、DeepSeek Harness(DSH)等任何 MCP 客户端直接增删改查论文。
**一句话用法**:AI 生成完整论文 → 调 `paper_write_text` 一次写入 → **一句话用法**:AI 生成完整论文 → 调 `paper_write_text` 一次写入 →
用 `paragraph_write` / `sentence_update` 微调 → `paper_document` 读回来检查。 用 `paragraph_write` / `sentence_update` 微调 → `paper_document` 读回来检查。
- 代码:`backend/app/mcp/`(工具实现)、`backend/scripts/mcp_server.py`(入口) - 代码:`backend/app/mcp/`(工具实现)、`backend/scripts/mcp_server.py`(入口)
- 冒烟测试:`backend/scripts/smoke_mcp.py`(真 MCP 客户端跑 76 项检查、覆盖 41/41 个工具,stdio 与 HTTP 都过) - 冒烟测试:`backend/scripts/smoke_mcp.py`(真 MCP 客户端跑 99 项检查、覆盖 41/41 个工具,stdio 与 HTTP 都过)
- 工具数量:**41 个**,按 `paper_` / `paragraph_` / `sentence_` / `template_` / `field_` / `reference_` / `reference_tag_` 七组前缀 - 工具数量:**41 个**,按 `paper_` / `paragraph_` / `sentence_` / `template_` / `field_` / `reference_` / `reference_tag_` 七组前缀
- 协议:MCP 2024-11-05 ~ 2025-11-25(官方 Python SDK `mcp>=2.2,<3` 协商) - 协议:MCP 2024-11-05 ~ 2025-11-25(官方 Python SDK `mcp>=2.2,<3` 协商)
- 传输:`stdio`(本地拉起)与 `streamable-http`(远程连接,支持 Bearer 口令),另兼容旧 `sse` - 传输:`stdio`(本地拉起)与 `streamable-http`(远程连接,支持 Bearer 口令),另兼容旧 `sse`
@@ -98,10 +98,12 @@ cd /data/project/paper/backend
.venv/bin/python scripts/smoke_mcp.py --url http://127.0.0.1:8931/mcp --token 你的口令 # 连已有的 HTTP 服务 .venv/bin/python scripts/smoke_mcp.py --url http://127.0.0.1:8931/mcp --token 你的口令 # 连已有的 HTTP 服务
``` ```
同一个套件 **76 项检查**跑两种传输,覆盖 21 个步骤:建论文 → 看骨架 → 按标题写段落 → 同一个套件 **99 项检查**跑两种传输,覆盖 23 个步骤:建论文 → 看骨架 → 按标题写段落 →
定位失败时列出候选标题 → 读段落/追加带引用的句子 → 只改一句 → 删一句 → 整篇一次写入 → 定位失败时列出候选标题 → 读段落/追加带引用的句子 → 只改一句 → 删一句 → 整篇一次写入 →
读全文 → 跨论文搜索 → 搬段落 → 清空段落 → 字段库增删改 → 建/改模板 → 换模板后正文不丢 → 读全文 → 跨论文搜索 → **另起一行**(公式独行、标记开关、搬段落时跟着走)→ **参考文献表**
文献库增删改查与批量导入 → 引用守卫(被引用的文献删不掉)→ 删模板被拒 → 删论文要确认 → 清理。写真实数据,结束时自己删干净。 (一处引用一行、一部文献一条、正文编号与表号同号)→ 搬段落 → 清空段落 → 字段库增删改 →
建/改模板 → 换模板后正文不丢 → 文献库增删改查与批量导入 → 引用守卫(被引用的文献删不掉)→
删模板被拒 → 删论文要确认 → 清理。写真实数据,结束时自己删干净。
--- ---
@@ -270,10 +272,10 @@ Header: Authorization: Bearer <MCP_HTTP_TOKEN>
| 工具 | 作用 | 关键参数 | | 工具 | 作用 | 关键参数 |
|---|---|---| |---|---|---|
| `paragraph_list` | 段落骨架 + 预览 | `paper_id`、`include_empty?` `preview?` | | `paragraph_list` | 段落骨架 + 预览 | `paper_id`、`include_empty?` `preview?` |
| `paragraph_get` | 读一段:拼好的正文、逐句 id/内容/引用/分隔符 | `paper_id`、`position?` `heading?` | | `paragraph_get` | 读一段:拼好的正文、逐句 id/内容/引用/分隔符与换行 | `paper_id`、`position?` `heading?` |
| `paragraph_write` | 写一段(覆盖或追加) | `paper_id`、`position?` `heading?`、`sentences?` `text?` `split?` `citations?` `mode?` | | `paragraph_write` | 写一段(覆盖或追加) | `paper_id`、`position?` `heading?`、`sentences?` `text?` `split?` `citations?` `mode?` |
| `paragraph_delete` | 清空该段全部句子(段落仍在) | `paper_id`、`position?` `heading?` | | `paragraph_delete` | 清空该段全部句子(段落仍在) | `paper_id`、`position?` `heading?` |
| `paragraph_move` | 整段搬到另一个位置(接在目标段之后) | `paper_id`、`position?` `heading?`、`target_position?` `target_heading?` | | `paragraph_move` | 整段搬到另一个位置(**接在目标段已有内容之后**,换行与引用一起搬过去) | `paper_id`、`position?` `heading?`、`target_position?` `target_heading?` |
定位规则:`position` 直接采用(包括模板没定义的位置,会以"未设定"渲染); 定位规则:`position` 直接采用(包括模板没定义的位置,会以"未设定"渲染);
`heading` 按模板解析,忽略编号、空格和大小写,所以下面三种写法等价: `heading` 按模板解析,忽略编号、空格和大小写,所以下面三种写法等价:
@@ -292,11 +294,38 @@ Header: Authorization: Bearer <MCP_HTTP_TOKEN>
| 工具 | 作用 | 关键参数 | | 工具 | 作用 | 关键参数 |
|---|---|---| |---|---|---|
| `sentence_list` | 列句子(可只列某段) | `paper_id`、`position?` `heading?` `include_empty?` | | `sentence_list` | 列句子(可只列某段) | `paper_id`、`position?` `heading?` `include_empty?` |
| `sentence_add` | 段尾追加一句(可带引用) | `paper_id`、`content`、`position?` `heading?` `citations?` `sort?` | | `sentence_add` | 段尾追加一句(可带引用) | `paper_id`、`content`、`position?` `heading?` `citations?` `line_break_before?` `sort?` |
| `sentence_update` | 改正文 / 改所属段 / 改排序 / 整组换引用 | `paper_id`、`sentence_id`、`content?` `position?` `sort?` `citations?` | | `sentence_update` | 改正文 / 改所属段 / 改排序 / 改换行 / 整组换引用 | `paper_id`、`sentence_id`、`content?` `position?` `sort?` `line_break_before?` `citations?` |
| `sentence_delete` | 删一句(连引用) | `paper_id`、`sentence_id` | | `sentence_delete` | 删一句(连引用) | `paper_id`、`sentence_id` |
| `sentence_search` | **跨论文搜句子**(保持术语一致) | `keyword`、`paper_id?` `limit?` | | `sentence_search` | **跨论文搜句子**(保持术语一致) | `keyword`、`paper_id?` `limit?` |
#### 另起一行:公式、分条列举、作者行
一段是**它的句子连排**出来的,所以"不是散文的行"要自己开口。方式是在句子对象上加
`line_break_before: true`,写进去、读回来、搬走都带着:
```jsonc
{"heading": "3. Methods", "sentences": [
"本段说明公式的来历。",
{"content": "$$ E = mc^2 $$"}, // 不用标,见下
{"content": "其中 $c$ 为光速。", "line_break_before": true},
{"content": "(1)第一项;", "line_break_before": true}
]}
```
三条规则:
1. **换行顶掉分隔符,不是叠加。** 新行本身已经把两句分开了,再加一个空格会让下一行空一格。
2. **display 公式(`$$…$$` 或 `\[…\]`)自己独占一行,不必标。** 论文页面就是这么排的,
`paper_document` 的 `text` 也这么排——两处必须一致,所以模型不要去猜"这里要不要加标记"。
行内公式(`$…$`)不算,它本来就在句子里。
3. **"请读一遍这一段"时看得到换行。** `paragraph_get` / `paper_document` 返回的 `text` 里换行就是真的换行符;
`sentence_list` 每句还带 `line_break_before` 字段,所以搬句子之前能看出哪几句是另起的行。
`text` + `split` 那种整块写法也可以带 `line_break_before`,但它只作用于**切出来的第一句**——
"这一块从新的一行开始"。块里其余几句是接着它的,不是各自另起一行;否则一个换行写了三行的段落
就会被切成三行。
### 3.4 模板 templates(5 个) ### 3.4 模板 templates(5 个)
模板就是结构:论文的标题是**每次渲染现读模板**得到的,所以改模板会影响所有用它的论文, 模板就是结构:论文的标题是**每次渲染现读模板**得到的,所以改模板会影响所有用它的论文,
@@ -418,6 +447,10 @@ Header: Authorization: Bearer <MCP_HTTP_TOKEN>
> `sentence` 会避开小数点(`3.14`)和 `et al.` / `i.e.` 这类缩写,但它仍然是"猜"。 > `sentence` 会避开小数点(`3.14`)和 `et al.` / `i.e.` 这类缩写,但它仍然是"猜"。
> 想完全不猜,就直接给 `sentences` 数组。 > 想完全不猜,就直接给 `sentences` 数组。
`paper_write_text` 的正文是一整块字符串,**表达不了"这一句另起一行"**(也只有第一行能带)。
display 公式不用担心——它本来就独占一行;但分条列举、作者行这类要另起的行,
请改用 `paper_write` 给 `sentences` 数组并加 `line_break_before: true`(见 §3.3)。
标题识别两种形态:Markdown 标题行(`# 0 Abstract`),或**与模板完全同名的独立一行** 标题识别两种形态:Markdown 标题行(`# 0 Abstract`),或**与模板完全同名的独立一行**
(把 `paper_outline` 的标题原样抄下来最稳)。前面没有标题的内容会被丢弃; (把 `paper_outline` 的标题原样抄下来最稳)。前面没有标题的内容会被丢弃;
一个都没匹配上则整次调用报错,不会写半截。 一个都没匹配上则整次调用报错,不会写半截。
@@ -565,6 +598,13 @@ REST 和 MCP 同时生效。这让"网页里看到的"和"AI 写进去的"永远
写入路径不切任何文本"(见 `docs/OVERVIEW.md`)。只有 `text` + `split` 这条路会切, 写入路径不切任何文本"(见 `docs/OVERVIEW.md`)。只有 `text` + `split` 这条路会切,
默认还是 `line`。已存文本永远不会被重新切分。 默认还是 `line`。已存文本永远不会被重新切分。
**换行是"要求"出来的,不是猜出来的。** `line_break_before` 只在被明确写出来时才设:
`text` 里换了一行不等于要另起一行(模型把一段话折成三行是常事),所以工具不从这里推断。
唯一的例外是 display 公式——页面渲染和 `paper_document` 的 `text` 都让它独占一行,
这条规则在 `frontend/src/utils/math.ts` 和 `app/mcp/support.py` 里各写了一遍
(一个 TypeScript、一个 Python),**改一处必须改另一处**,否则网页和工具结果就会在
"哪一行是哪一行"上不一致。冒烟测试里有一项专门盯着它。
**删除要二次确认。** `paper_delete` 第一次调用必定被拒并说明后果, **删除要二次确认。** `paper_delete` 第一次调用必定被拒并说明后果,
因为一次误删会连带整篇的句子和引用,而这在 MCP 里没有回收站。 因为一次误删会连带整篇的句子和引用,而这在 MCP 里没有回收站。
+2 -1
View File
@@ -516,7 +516,7 @@ written by an agent is indistinguishable from one written by hand. What
| Addition | Why the REST shape is wrong for a model | | Addition | Why the REST shape is wrong for a model |
|---|---| |---|---|
| 29 tools with `paper_` / `paragraph_` / `sentence_` / `template_` / `field_` prefixes | a model picks a tool out of a list by its name, not by reading 29 descriptions | | 41 tools with `paper_` / `paragraph_` / `sentence_` / `template_` / `field_` / `reference_` / `reference_tag_` prefixes | a model picks a tool out of a list by its name, not by reading 41 descriptions |
| JSON strings, `None`-free, no pagination envelope on reads | a tool result is paid for in context tokens | | JSON strings, `None`-free, no pagination envelope on reads | a tool result is paid for in context tokens |
| `paragraph_write` addressed by **heading** as well as position | nobody writing "1. Introduction" knows the template places it at `sort = 20` | | `paragraph_write` addressed by **heading** as well as position | nobody writing "1. Introduction" knows the template places it at `sort = 20` |
| `paper_write` / `paper_write_text` | "generate the paper, then put it in" is one intention, not thirty round trips | | `paper_write` / `paper_write_text` | "generate the paper, then put it in" is one intention, not thirty round trips |
@@ -524,6 +524,7 @@ written by an agent is indistinguishable from one written by hand. What
| `sentence_search` across papers | the job is consistency — one paper says 洪水损失, the next must not say GUL | | `sentence_search` across papers | the job is consistency — one paper says 洪水损失, the next must not say GUL |
| `reference_*` and `reference_tag_*` | the reference library is what makes a citation mean something; an agent that can cite id 42 but cannot look up what 42 *is* is writing blind | | `reference_*` and `reference_tag_*` | the reference library is what makes a citation mean something; an agent that can cite id 42 but cannot look up what 42 *is* is writing blind |
| `reference_import` | "here are the twenty references I cited" is one intention, and twenty round trips would spend twenty schemas on it | | `reference_import` | "here are the twenty references I cited" is one intention, and twenty round trips would spend twenty schemas on it |
| `line_break_before` on a sentence, and the display-formula rule in `document_text` | a paragraph is its sentences run together, so a formula, a list or an author line needs a way to say where the line starts — and the text an agent reads back has to break lines where the paper view breaks them, or the two disagree about what the manuscript looks like |
Two transports are served from one build: `stdio`, which is what a client Two transports are served from one build: `stdio`, which is what a client
spawns, and `streamable-http`, which is what a client on another machine spawns, and `streamable-http`, which is what a client on another machine