能力标签

🔌 MCP 🤖 Agent 🔄 工作流 🌐 翻译 🐳 Docker 🔗 REST API 🧬 Embedding 🧠 Claude ✨ GPT ⛓ LangChain

🔌

MCP工具

证据实验室

基于 Python · 让 AI 助手直接操作你的系统与工具

英文名：evidencelab

⭐ 33 Stars 🍴 6 Forks 💻 Python 📄 MIT 🏷 AI 7.5分

7.5AI 综合评分

mcppython文档流水线

⬇ 下载源码 ZIP ⚙️ 配置说明

✦ AI Skill Hub 推荐

AI Skill Hub 推荐使用：证据实验室是一款优质的MCP工具。AI 综合评分 7.5 分，在同类工具中表现稳健。如果你正在寻找可靠的MCP工具解决方案，这是一个值得深入了解的选择。

📚 深度解析

证据实验室是一款基于 MCP（Model Context Protocol）标准协议的 AI 工具扩展。MCP 协议由 Anthropic 开发并开源，旨在建立 AI 模型与外部工具之间的标准化通信接口，目前已被 Claude Desktop、Claude Code、Cursor 等主流 AI 工具采纳。

通过安装证据实验室，你的 AI 助手将获得额外的工具调用能力，可以用自然语言直接操控该工具的功能，无需学习复杂的命令行语法。MCP 工具的核心价值在于"一次配置，永久增强"——配置完成后，每次与 AI 对话时都可以无缝调用这些工具。

在技术实现上，MCP 工具通过标准的 JSON-RPC 协议与 AI 客户端通信，工具的功能以"工具列表"的形式暴露给 AI 模型，AI 可以按需调用。证据实验室提供了结构化的工具调用接口，使 AI 模型能够精确地理解和使用每个功能点，显著降低 AI 在工具使用上的错误率。

与传统的 API 集成相比，MCP 工具的优势在于无需编写代码——用户只需在配置文件中添加几行 JSON，即可让 AI 获得全新能力。AI Skill Hub 将证据实验室评为 AI 评分 7.5 分，属于同类工具中的优质选择。

📋 工具概览

证据实验室是一款遵循 MCP（Model Context Protocol）标准协议的 AI 工具扩展。通过 MCP 协议，它可以让 Claude、Cursor 等主流 AI 客户端直接访问和操作外部工具、数据源和服务，实现 AI 能力的无缝扩展。无论是文件操作、数据库查询还是 API 调用，都可以通过自然语言在 AI 对话中直接触发，极大提升生产效率。

GitHub Stars

⭐ 33

开发语言

Python

支持平台

Windows / macOS / Linux

维护状态

轻量级项目，按需更新

开源协议

MIT

AI 综合评分

7.5 分

工具类型

MCP工具

Forks

📖 中文文档

以下内容由 AI Skill Hub 根据项目信息自动整理，如需查看完整原始文档请访问底部「原始来源」。

📌 核心特色

通过标准 MCP 协议与 Claude、Cursor 等主流 AI 客户端深度集成
提供结构化工具调用接口，显著降低 AI 集成复杂度
支持 Claude Desktop 和 Claude Code 无缝接入，开箱即用
可与其他 MCP 工具组合叠加，构建完整 AI 工作站
轻量无侵入设计，不影响现有系统架构

🎯 主要使用场景

在 Claude Desktop 对话中直接调用本地工具，实现 AI 与系统的深度联动
通过自然语言驱动复杂的多步骤自动化任务，代替繁琐手动操作
将多个 MCP 工具组合使用，构建个人专属 AI 工作站

以下安装命令基于项目开发语言和类型自动生成，实际以官方 README 为准。

安装命令

# 方式一：通过 Claude Code CLI 一键安装
claude skill install https://github.com/dividor/evidencelab

# 方式二：手动配置 claude_desktop_config.json
{
  "mcpServers": {
    "-----": {
      "command": "npx",
      "args": ["-y", "evidencelab"]
    }
  }
}

# 配置文件位置
# macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
# Windows: %APPDATA%/Claude/claude_desktop_config.json

📋 安装步骤说明

确认已安装 Node.js（v18 或以上版本）
打开 Claude Desktop 或 Claude Code 的 MCP 配置文件
按「交给 Agent 安装 → Claude Desktop」标签中的 JSON 配置填入 mcpServers 字段
保存配置文件并重启 Claude 客户端
重启后，在对话中即可使用本工具

以下用法示例由 AI Skill Hub 整理，涵盖最常见的使用场景。

常用命令 / 代码示例

# 安装后在 Claude 对话中直接使用
# 示例：
用户: 请帮我用 证据实验室 执行以下任务...
Claude: [自动调用 证据实验室 MCP 工具处理请求]

# 查看可用工具列表
# 在 Claude 中输入："列出所有可用的 MCP 工具"

以下配置示例基于典型使用场景生成，具体参数请参照官方文档调整。

配置示例

// claude_desktop_config.json 配置示例
{
  "mcpServers": {
    "_____": {
      "command": "npx",
      "args": ["-y", "evidencelab"],
      "env": {
        // "API_KEY": "your-api-key-here"
      }
    }
  }
}

// 保存后重启 Claude Desktop 生效

📑 README 深度解析真实文档完整度 69/100 查看 GitHub 原文 →

以下内容由系统直接从 GitHub README 解析整理，保留代码块、表格与列表结构。

Evidence Lab

📖 Comprehensive documentation is available at evidencelab.ai/docs, sourced from the docs/ folder in this repository.

Introduction

Evidence Lab

Evidence Lab is a free open source platform that provides a document pipeline, search, and AI-powered information discovery tools. The aim is to provide a quick start for those looking to use AI with their documents and a place where new ideas can be tested.

You can run the code yourself, or explore the online version at evidencelab.ai which has so far been populated with about 20,000 United Nations humanitarian evaluation reports sourced from the United Nations Evaluation Group. See Data for more information on these amazing documents.

If you would like to have your public documents added to Evidence Lab, or would like to contribute to the project, please reach out to evidencelab@astrobagel.com.

Also, for the latest news check out the AstroBagel Blog.

Features

Evidence Lab document processing pipeline includes the following features:

Processing pipeline

PDF/Word parsing with Docling, to include document structure detection
Footnote and references, images and table detection
Basic table extraction, with support for more expensive processing as required
AI-assisted document summarization
AI-assisted tagging of documents
Indexing with Open (Huggingface) or proprietary models (Azure foundry, but extensible)

User interface

Search	Research Assistant	Heatmapper	Pipeline
<img src="docs/images/search-guide/search-results-full.png" alt="Search" height="200">	<img src="docs/images/assistant/assistant-response.png" alt="Research Assistant" height="200">	<img src="docs/images/heatmapper-overview.png" alt="Heatmapper" height="200">	<img src="docs/images/monitor/pipeline-view.png" alt="Pipeline" height="200">

Hybrid search with AI summary and reranking
Research Assistant — chat-based AI agent that searches, analyzes, and synthesizes findings with inline citations and multi-turn conversations with thread history
Deep Research mode — coordinator/researcher sub-agent architecture using deepagents for thorough multi-step investigations with real-time streaming progress
Star ratings — rate search results, AI summaries, and assistant responses with 1–5 stars and optional comments
Drilldown research — highlight text or click "Find out more" to drill into sub-topics, building an explorable research tree with query inheritance and PDF export
Field boosting — detects countries/organizations in the query and promotes matching results; at full weight, non-matching results are excluded
Experimental features such as heatmapper for tracking trends in content
Config-driven filter fields — control which metadata fields appear in the filter panel
Filtering by metadata, in-document section types
Search and reranking settings to explore different models
Auto min score filtering using percentile-based thresholding (filters bottom 30% of results)
Semantic highlighting in search results
Basic language translation
PDF preview with in-document search
Built-in searchable documentation area with sidebar navigation
Administration views to track pipeline, documents, performance and errors

AI platform integrations

REST API — full programmatic access to search, documents, and admin functions. Protected by API key authentication. Interactive docs at /api/docs (Swagger UI). See API docs
MCP Server — Model Context Protocol server allowing Claude, ChatGPT, and other AI assistants to search Evidence Lab documents as tools. Connect via + > Connectors in Claude or + > More Add Sources in ChatGPT. Tools: search (semantic search with filters, facets, and citations) and get_document (full document metadata)
A2A Agent — Agent-to-Agent protocol server for AI agent frameworks (Google ADK, CrewAI, LangGraph, etc.). Exposes a research skill (full assistant synthesis with streaming) and search skill. Runs on the same service as MCP. Agent Card at /.well-known/agent.json
Both protocols share OAuth 2.0 authentication (or X-API-Key header), rate limiting, and audit logging
API keys are managed via Admin → API Keys. After generating a new key, allow up to 60 seconds before using it with MCP or A2A (key hashes are cached per-process on a 60-second TTL). The API_SECRET_KEY env var takes effect immediately.

User authentication & permissions (opt-in)

Email/password registration with email verification, or OAuth single sign-on (Google, Microsoft)
Cookie-based sessions with CSRF protection — no tokens in localStorage
Account lockout, rate limiting, and audit logging for security
Group-based data-source access control — restrict which datasets users can see
Admin panel for managing users, groups, and permissions
User feedback — rate search results, AI summaries, documents, and taxonomy with 1–5 stars
Activity logging — automatic search activity capture with admin views and XLSX export
Self-service profile management and account deletion
Built on fastapi-users with future MFA support in mind
Three modes via USER_MODULE in .env: off (default), on_passive (optional login), on_active (login required)

See the CHANGELOG for the full list of recent additions.

Getting started

You can explore the hosted version at evidencelab.ai.

Run the demo — interactive setup will prompt for provider and API keys

python scripts/demo/run_demo.py --mode host


The script will automatically configure `.env`, add a demo datasource to
`config.json`, download documents, and run the pipeline.

**Running in Docker** (guaranteed to work on any Docker-capable machine, but
can be significantly slower as it cannot utilise GPU or Apple MPS acceleration
on your host):

bash

Quick Start

1. Configure data sources - Edit config.json in the repo root to define datasources, data_subdir, field_mapping, and taxonomies. - The UI reads the same config.json via Docker Compose.

2. Set environment variables - Copy .env.example to .env. - Fill in the API keys and service URLs required by the pipeline and UI.

3. Add documents + metadata - Save documents under data/<data_subdir>/pdfs/<organization>/<year>/. - For each document, include a JSON metadata file with the same base name. - If a download failed, add a .error file with the same base name (scanner records these).

Example layout:

   data/
     uneg/
       pdfs/
         UNDP/
           2024/
             report_123.pdf
             report_123.json
             report_124.error (if there was an error downloading the file)

4. Run the pipeline (Docker)

   # Start services
   docker compose up -d --build

   # Run the orchestrator (example: UNEG)
   docker compose exec pipeline \
     python -m pipeline.orchestrator --data-source uneg --skip-download --num-records 10

Tip: To quickly ingest a single test document and verify the full stack, > run the integration test script instead: >
   > ./tests/integration/run_integration_host_pipeline.sh
   > 
> This ingests a sample report, rebuilds the containers, and runs the > integration test suite end-to-end.

5. Access the Evidence Lab UI - Open http://localhost:3000 - Select your data source and search the indexed documents

6. Next steps - To add user authentication see User authentication below - See the technical deep dive for pipeline commands, downloaders, and architecture details: ui/frontend/public/docs/tech.md - See CONTRIBUTING.md for development setup, pre-commit hooks, testing, and contribution guidelines

Demo (quickest way to try it)

The interactive demo script guides you through provider selection, API key setup, downloads a few World Bank documents, and runs the full pipeline.

Running on host (recommended — can use hardware acceleration such as Apple MPS or NVIDIA CUDA, but may require some adjustments to suit your environment):

```bash

Run the demo

python scripts/demo/run_demo.py --mode docker


Once complete, open http://localhost:3000 and select the **demo** data source.

**Options:**

bash python scripts/demo/run_demo.py --mode host --num-docs 10 # Download more documents python scripts/demo/run_demo.py --mode host --skip-download # Re-run pipeline only python scripts/demo/run_demo.py --mode host --skip-pipeline # Download only ```

Create and activate a virtual environment

python3 -m venv ~/.venvs/evidencelab-ai source ~/.venvs/evidencelab-ai/bin/activate pip install -r requirements.txt

Configuration Reference

All configuration lives in a single config.json at the repo root. The file is shared between the pipeline and the UI via Docker Compose volumes.

2. Configure email (SMTP)

Email is used for account verification and password resets. For production, configure a real SMTP provider (SendGrid, AWS SES, Gmail, etc.):

SMTP_HOST=smtp.example.com
SMTP_PORT=587
SMTP_USER=apikey
SMTP_PASSWORD=your-smtp-password
SMTP_FROM=noreply@yourdomain.com
SMTP_USE_TLS=true

For local development, use Mailpit — a lightweight SMTP server that catches all outgoing emails:

```bash

3. Configure OAuth (optional)

To enable Google and/or Microsoft single sign-on, add the relevant credentials to .env:

```env

5. Configure groups and data-source access

Evidence Lab uses groups to control which data sources users can see:

A Default group is created automatically and grants access to all data sources. New users are added to this group on registration.
To restrict access, create additional groups from the Admin → Groups panel, assign specific data-source keys to each group, and move users into the appropriate groups.
Users who are only in non-default groups will see only the data sources assigned to their groups.

Additional settings

See .env.example for the full list of auth-related settings including:

Setting	Default	Description
`FIRST_SUPERUSER_EMAIL`	(empty)	Email of the account to auto-promote to admin on startup
`AUTH_ALLOWED_EMAIL_DOMAINS`	(empty — open)	Comma-separated whitelist of allowed email domains
`AUTH_MIN_PASSWORD_LENGTH`	`8`	Minimum password length
`AUTH_COOKIE_SECURE`	`true`	Set to `false` for non-HTTPS local dev
`AUTH_RATE_LIMIT_MAX`	`10`	Max login attempts per IP per window
`AUTH_RATE_LIMIT_WINDOW`	`60`	Rate limit window in seconds
`AUTH_LOCKOUT_THRESHOLD`	`5`	Failed logins before account lockout
`AUTH_LOCKOUT_DURATION_MINUTES`	`15`	Lockout duration

1. Enable the module

USER_MODULE supports three modes:

Mode	Description
`off`	No authentication (default)
`on_passive`	Auth UI available but optional — anonymous users can browse freely, registered users get profiles and permissions
`on_active`	All access requires login — unauthenticated users cannot see datasources

Set these in your .env:

USER_MODULE=on_active
REACT_APP_USER_MODULE=on_active
AUTH_SECRET_KEY=<generate-a-random-secret-at-least-32-characters>

Legacy values true/false are still supported (true → on_active, false → off).

Tip: Generate a secret with python -c "import secrets; print(secrets.token_urlsafe(32))".

🎯 aiskill88 AI 点评 A 级 2026-06-11

高质量的开源MCP工具，值得关注

📚 实用指南（长尾问题）

适合谁

需要让 Claude / Cursor 操作本地工具的 AI 工程师

最佳实践

配置 MCP 服务器时建议使用 stdio 传输 + JSON-RPC，避免暴露公网

常见错误

API key 直接提交到 git 仓库（请用 .env 并加入 .gitignore）
MCP 配置路径拼错或权限不足，重启 Claude Desktop 才生效
Python 依赖冲突：建议用 venv / uv 隔离环境

部署方案

云端托管：可放在 Vercel / Railway / Fly.io 等 PaaS 平台

⚡ 核心功能

通过标准 MCP 协议与 Claude、Cursor 等主流 AI 客户端深度集成
提供结构化工具调用接口，显著降低 AI 集成复杂度
支持 Claude Desktop 和 Claude Code 无缝接入，开箱即用
可与其他 MCP 工具组合叠加，构建完整 AI 工作站
轻量无侵入设计，不影响现有系统架构

👥 适合谁

需要让 Claude / Cursor 操作本地工具的 AI 工程师

⭐ 最佳实践

配置 MCP 服务器时建议使用 stdio 传输 + JSON-RPC，避免暴露公网

⚠️ 常见错误

API key 直接提交到 git 仓库（请用 .env 并加入 .gitignore）
MCP 配置路径拼错或权限不足，重启 Claude Desktop 才生效
Python 依赖冲突：建议用 venv / uv 隔离环境

👥 适合人群

Claude Desktop / Claude Code 用户AI 工具开发者需要扩展 AI 能力的专业人士自动化工程师

🎯 使用场景

在 Claude Desktop 对话中直接调用本地工具，实现 AI 与系统的深度联动
通过自然语言驱动复杂的多步骤自动化任务，代替繁琐手动操作
将多个 MCP 工具组合使用，构建个人专属 AI 工作站

⚖️ 优点与不足

✅ 优点

+MIT 协议，可免费商用
+标准化 MCP 协议，生态互联性强
+与 Claude 官方生态无缝对接
+即插即用，配置简单快捷

⚠️ 不足

−依赖 Claude 客户端，非 Claude 用户无法使用
−MCP 协议仍在持续演进，接口可能变更
−需要一定的配置步骤

⚠️ 使用须知

AI Skill Hub 为第三方内容聚合平台，本页面信息基于公开数据整理，不对工具功能和质量作任何法律背书。

建议在沙箱或测试环境中充分验证后，再部署至生产环境，并做好必要的安全评估。

📄 License 说明

🔗 相关工具推荐

🧩 你可能还需要

基于当前 Skill 的能力图谱，自动补全的工具组合

natively-cluely-ai-assistant — Claude Skill 中文使用文档

免费开源的AI面试助手，实时转录，隐蔽模式，局部RAG，BYOK。无订阅，防止数据泄露。

❓ 常见问题 FAQ

如何使用证据实验室？−

请参考项目文档和示例代码

MCP 是什么？和普通 API 有什么区别？+

我需要编程基础才能使用这个 MCP 工具吗？+

这个工具支持 Claude Code 吗？还是只有 Claude Desktop？+

MCP 工具是否会访问我的本地文件或数据？+

多个 MCP 工具可以同时使用吗？+

💡 AI Skill Hub 点评

总体来看，证据实验室是一款质量良好的MCP工具，在同类工具中具备一定竞争力。AI Skill Hub 将持续追踪其更新动态，建议收藏备用，结合自身场景选择合适时机引入使用。

⬇️ 获取与下载

⬇ 下载源码 ZIP

✅ MIT 协议 · 可免费商用 · 直接从 aiskill88 服务器下载，无需跳转 GitHub

📚 深入学习证据实验室

查看分步骤安装教程和完整使用指南，快速上手这款工具

⚙️ 安装教程 📚 使用教程

🌐 原始信息

原始名称	`evidencelab`
原始描述	开源MCP工具：Evidence lab is a free open source platform that provides a document pipeline, s。⭐33 · Python
Topics	`mcppython文档流水线`
GitHub	https://github.com/dividor/evidencelab
License	MIT
语言	Python

🔗 原始来源

🐙 GitHub 仓库 https://github.com/dividor/evidencelab

收录时间：2026-06-11 · 更新时间：2026-06-11 · License：MIT · AI Skill Hub 不对第三方内容的准确性作法律背书。

证据实验室

📚 深度解析

📋 工具概览

📖 中文文档

Evidence Lab

Introduction

Features

Getting started

Run the demo — interactive setup will prompt for provider and API keys

Quick Start

Demo (quickest way to try it)

Run the demo

Create and activate a virtual environment

Configuration Reference

2. Configure email (SMTP)

3. Configure OAuth (optional)

5. Configure groups and data-source access

Additional settings

1. Enable the module

⚡ 核心功能

👥 适合人群

🎯 使用场景

⚖️ 优点与不足

🔗 相关工具推荐

❓ 常见问题 FAQ

🤖 交给 Agent 安装 · 证据实验室