collect_evidence

Search academic literature for evidence on a specific topic, optionally focusing on particular sections like methodology or findings, to support research and analysis.

Instructions

收集特定主题的文献证据

搜索与主题相关的文献片段，可选择聚焦于特定章节类型。

Args: topic: 搜索主题 section_focus: 聚焦的章节类型（如 "methodology", "findings"） k: 返回结果数量

Returns: 按文献聚合的证据列表

Input Schema

TableJSON Schema

Name	Required	Description	Default
`topic`	Yes
`section_focus`	No
`k`	No

Implementation Reference

src/paperlib_mcp/tools/writing.py:802-885 (handler)

The core handler implementation for the 'collect_evidence' tool. It adjusts the query based on optional section_focus, performs a hybrid search using hybrid_search, aggregates evidence chunks by document with metadata, sorts by evidence count per doc, and returns a structured dictionary.

@mcp.tool()
async def collect_evidence(
    topic: str,
    section_focus: str | None = None,
    k: int = 20,
) -> dict[str, Any]:
    """收集特定主题的文献证据
    
    搜索与主题相关的文献片段，可选择聚焦于特定章节类型。
    
    Args:
        topic: 搜索主题
        section_focus: 聚焦的章节类型（如 "methodology", "findings"）
        k: 返回结果数量
        
    Returns:
        按文献聚合的证据列表
    """
    try:
        # 如果有章节聚焦，调整查询
        query = topic
        if section_focus:
            focus_keywords = {
                "methodology": "method approach model estimation identification",
                "findings": "result finding evidence show demonstrate",
                "theory": "theory framework hypothesis prediction",
                "data": "data sample variable measure",
            }
            if section_focus in focus_keywords:
                query = f"{topic} {focus_keywords[section_focus]}"
        
        # 搜索
        search_result = await hybrid_search(query, k=k, alpha=0.6, per_doc_limit=5)
        
        # 按文档聚合
        evidence_by_doc: dict[str, dict] = {}
        
        for result in search_result.results:
            doc_id = result.doc_id
            
            if doc_id not in evidence_by_doc:
                # 获取文档信息
                doc = query_one(
                    "SELECT title, authors, year FROM documents WHERE doc_id = %s",
                    (doc_id,)
                )
                evidence_by_doc[doc_id] = {
                    "doc_id": doc_id,
                    "title": doc["title"] if doc else "Unknown",
                    "authors": doc["authors"] if doc else "Unknown",
                    "year": doc["year"] if doc else None,
                    "evidence": [],
                }
            
            evidence_by_doc[doc_id]["evidence"].append({
                "chunk_id": result.chunk_id,
                "page_start": result.page_start,
                "page_end": result.page_end,
                "text": result.snippet,
                "relevance_score": result.score_total,
            })
        
        # 按证据数量排序
        sorted_evidence = sorted(
            evidence_by_doc.values(),
            key=lambda x: len(x["evidence"]),
            reverse=True
        )
        
        return {
            "topic": topic,
            "section_focus": section_focus,
            "total_chunks": len(search_result.results),
            "unique_documents": len(sorted_evidence),
            "evidence": sorted_evidence,
        }
        
    except Exception as e:
        return {
            "error": str(e),
            "topic": topic,
            "evidence": [],
        }

src/paperlib_mcp/server.py:37-37 (registration)
Registers the writing tools module (including collect_evidence) by calling register_writing_tools on the FastMCP instance in the main server file.
```
register_writing_tools(mcp)
```
src/paperlib_mcp/tools/writing.py:802-802 (registration)
The @mcp.tool() decorator directly registers the collect_evidence function as an MCP tool within the register_writing_tools function.
```
@mcp.tool()
```

Paperlib MCP