Skip to main content
Glama

📚 LiveDocs MCP

Give your AI the docs it actually needs — fresh from the source, every time.

Python FastMCP License Libraries Deploy


🔎 The Problem

Every AI engineer has hit this wall:

"Claude is giving me the LangChain v0.1 API but we're on v0.3" "Why is the OpenAI example broken — oh, they changed the client in v1.0" "The CrewAI docs changed again, my code is outdated"

LLMs are trained on a snapshot of the internet. Documentation moves fast. Your AI assistant doesn't.

LiveDocs MCP bridges that gap. It connects Claude (or any MCP-compatible AI) directly to official documentation websites in real time — no stale training data, no hallucinated APIs.


Related MCP server: Context7 MCP

✨ Features

Feature

Description

🔴Live Docs

Fetches directly from official docs — always current

📦45+ Libraries

Every major AI engineering library covered

⚡Smart Cache

24-hour TTL cache — fast repeat queries, no wasted API calls

🔁Multi-Library Search

Compare docs across frameworks side-by-side

💻Code Extraction

Hunts for code-block-rich pages automatically

🔢Version-Aware

Pin to a specific version (e.g.v0.2, 2.1.0)

🔑BYOK

Bring Your Own Serper API key per request

🌐Remote-Ready

Deploy as HTTP/SSE server — share one URL with your whole team


📦 Supported Libraries (45+)

Key

Library

Description

openai

OpenAI

GPT-4o, embeddings, DALL·E, Whisper

anthropic

Anthropic

Claude API — messages, tool use, vision

groq

Groq

Ultra-fast LLM inference

mistral

Mistral AI

Mixtral, Mistral-7B

cohere

Cohere

Command R, Embed, Rerank

together

Together AI

Open-source model inference

replicate

Replicate

Run open-source models via API

google_ai

Google Gemini

Gemini Pro, Flash, embeddings

langchain · langgraph · langsmith · llama_index · haystack · dspy

crewai · autogen · semantic_kernel · smolagents · pydantic_ai

pinecone · weaviate · chromadb · qdrant · milvus · pgvector

huggingface · sentence_transformers

langsmith · weights_biases · mlflow

fastapi · pydantic · django · flask

sqlalchemy · supabase · prisma · redis

uv · ruff · pytest · docker · pandas · polars


🛠️ MCP Tools Reference

Tool

Description

get_docs(query, library, version?, serper_api_key?)

Fetch live docs for any library

search_docs_multi(query, libraries[])

Search across multiple libraries at once

get_code_examples(query, library, language?)

Targeted code snippet extraction

list_libraries(category?)

Browse all 45+ supported libraries

cache_stats()

View cache hits / entries / TTL


🚀 Quick Start

Add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "live-docs": {
      "url": "https://live-docs-mcp.onrender.com/sse"
    }
  }
}

Config file location:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

  • Windows: %APPDATA%\Claude\claude_desktop_config.json


Option 2 — Local (Self-Hosted)

Prerequisites: Python 3.11+, uv

# 1. Clone
git clone https://github.com/NoumanZahid-85/live-docs-mcp.git
cd live-docs-mcp

# 2. Install dependencies
uv sync

# 3. Configure
cp .env.example .env
# Edit .env and add your SERPER_API_KEY (free at https://serper.dev)

# 4. Test the CLI client
uv run python mcp_client.py
# > Library: openai
# > Query: streaming chat completion

Claude Desktop config (local stdio):

{
  "mcpServers": {
    "live-docs": {
      "command": "uv",
      "args": ["run", "python", "mcp_server.py"],
      "cwd": "/absolute/path/to/live-docs-mcp",
      "env": {
        "SERPER_API_KEY": "your_serper_api_key"
      }
    }
  }
}

🌐 Deploy Your Own Instance

Render (Free Tier)

Deploy to Render

  1. Fork this repo

  2. Connect to Render → New Web Service → select your fork

  3. Set environment variable SERPER_API_KEY in the Render dashboard

  4. Render auto-detects render.yaml and deploys

Your server will be live at https://your-service.onrender.com/sse.


🏗️ Architecture

graph TD
    A[Claude / AI Client] -->|MCP Protocol| B[LiveDocs MCP Server]

    B --> C{Cache Hit?}
    C -->|Yes - 24h TTL| D[Return Cached Result]
    C -->|No| E[Serper Google Search API]

    E --> F[Top 5 Doc URLs]
    F --> G[Concurrent HTTP Fetch]
    G --> H[trafilatura HTML → Text]
    H --> I[Code Block Extraction]
    I --> J[Cache + Return]

    B --> K[list_libraries]
    B --> L[cache_stats]

    subgraph Tools
        B --> M[get_docs]
        B --> N[search_docs_multi]
        B --> O[get_code_examples]
    end

🔑 API Keys

Key

Purpose

Free Tier

Get It

SERPER_API_KEY

Google search via Serper

✅ 2,500 searches/month

serper.dev

GROQ_API_KEY

LLM answers in CLI client

✅ Generous free tier

console.groq.com


💡 Example Prompts

Once connected in Claude, try:

"Search the LangChain docs for how to use structured output with tool calling"

"Compare how LangChain and LlamaIndex implement document chunking"

"Get me a code example for streaming with the OpenAI Assistants API"

"What changed in CrewAI v0.7?"

"List all supported vector database libraries"

🗂️ Project Structure

live-docs-mcp/
├── mcp_server.py      # 🚀 Production MCP server (5 tools)
├── mcp_client.py      # 🖥️ Interactive CLI client
├── libraries.py       # 📋 Registry of 45+ libraries
├── cache.py           # ⚡ TTL in-memory cache
├── utils.py           # 🔧 HTML extractor + code parser
├── render.yaml        # 🌐 One-click Render deployment
├── pyproject.toml     # 📦 Dependencies (uv)
├── .env.example       # 🔑 Environment variable template
└── README.md

🤝 Contributing

Want to add a library? Edit libraries.py — it takes 5 lines:

"your_library": {
    "site": "docs.your-library.com",
    "description": "Short description of the library",
    "category": "llm_frameworks",   # see existing categories
    "aliases": ["yl", "yourlib"],   # optional short names
},

Then open a PR. ✨


📄 License

MIT — free to use, modify, and deploy.


Available Tools

5 tools
cache_statsCache StatsA

Return current cache statistics for the LiveDocs MCP server.

Useful for understanding cache hit rates and diagnosing repeated query latency.

Returns: Cache statistics as formatted markdown.

Inputs: no named parameters. Returns: result

get_code_examplesGet Code ExamplesA

Fetch working code examples for a specific library feature from official docs.

Unlike get_docs(), this tool specifically hunts for pages rich in code samples, extracting and formatting fenced code blocks for immediate use.

Inputs: query (required), library (required), language, serper_api_key. Returns: result

get_docsGet DocsA

Fetch real-time, up-to-date documentation from the official docs site of a library.

Solves the core AI-engineer problem: LLMs have outdated training data. This tool fetches live content directly from official documentation websites, so you always get the latest API signatures, parameters, and examples.

Inputs: query (required), library (required), version, serper_api_key. Returns: result

list_librariesList LibrariesA

List all 45+ supported documentation libraries, optionally filtered by category.

Inputs: category. Returns: result

search_docs_multiSearch Docs MultiA

Search the same query across multiple libraries simultaneously.

Ideal for comparing how different frameworks handle the same concept:

  • "How does LangChain vs LlamaIndex implement RAG?"

  • "Compare asyncio support in FastAPI vs Django"

  • "Tool calling: OpenAI vs Anthropic vs Groq"

Inputs: query (required), libraries (required), serper_api_key. Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.1/5.0

    Scored across 5 tools

    Disambiguation4/5

    Each tool has a generally clear role: single-library docs, multi-library search, code examples, library listing, and cache stats. The only minor overlap is between get_docs and get_code_examples, but the descriptions explicitly differentiate them by focusing on prose/docs versus code-rich pages.

    Naming Consistency4/5

    Most tools follow a verb_noun pattern: get_docs, search_docs_multi, get_code_examples, list_libraries. cache_stats breaks the pattern by being noun_noun, and search_docs_multi's suffix is slightly irregular, but overall the naming is predictable and readable.

    Tool Count5/5

    Five tools is well-scoped for a documentation-fetching server. Each tool serves a distinct and useful purpose without redundancy or bloat, and the count feels appropriate for the narrow but valuable domain.

    Completeness5/5

    The tool surface covers the core documentation workflows: fetching live docs, searching across libraries, retrieving code examples, discovering supported libraries, and inspecting cache health. No obvious dead ends or missing critical operations for the stated purpose.

    Maintenance

    ActivityStale
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers