EvidenceLens MCP
Enables deterministic, read-only multimodal evidence review by normalizing text, tables, PDFs, images, and screenshots into provenance metadata, with optional controlled filesystem access and no provider calls.
README
EvidenceLens MCP
EvidenceLens MCP is a TypeScript Model Context Protocol server for deterministic, read-only evidence review. The single review_evidence tool accepts four distinct required course roles plus optional evidence, normalizes inline and explicitly configured filesystem sources, and returns schema-validated findings with typed provenance. It remains provider-independent and makes no network or provider calls.
Local Development
npm install
npm run dev
npm test
npm run build
npm run dev starts the MCP server over stdio from src/server.ts. npm test runs the full contract, normalizer, fixture, safety, and MCP protocol test suite. npm run build type-checks and compiles the server.
MCP Contract
See docs/mcp-contract.md for the exact four-role request contract, duplicate-ID and stable errors, deterministic finding fields and citation mapping, root configuration grammar, byte limits, filesystem:// provenance, transient analysis boundary, and read-only/no-provider/no-Docker behavior.
Platform note: the default filesystem reader uses descriptor-relative component walking on Linux. On macOS, where this project has no supported Node openat/openat2 binding, default anchored filesystem reads fail closed with sanitized ACCESS_DENIED and no bytes; there is no pathname fallback. Phase 2 inline evidence remains supported, and embedding tests may inject a reviewed safe filesystem adapter for portable filesystem-read coverage.
Optional filesystem roots
Filesystem access is disabled when no roots are configured; the server never falls back to the current directory, home directory, repository, or another broad default. Configure explicit roots with repeated id=absolute-path entries separated by commas or semicolons:
EVIDENCELENS_ALLOWED_ROOTS='course=/absolute/path/course;examples=/absolute/path/examples' npm run dev
Root IDs match [A-Za-z][A-Za-z0-9_-]{0,31}. Empty entries, duplicate IDs, canonical path collisions, unreadable roots, non-directories, relative paths, and malformed entries fail with the stable message Invalid filesystem root configuration; paths are never echoed.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。