agents-docs
Ultra-fast local markdown documentation RAG for AI coding agents, enabling instant BM25 search over official framework docs and live AI model registries via MCP.
README
agents-docs
<p align="left"> <a href="https://github.com/Lolaplex/agents-docs/releases"><img src="https://img.shields.io/badge/version-0.42.0-blue.svg?style=flat-square" alt="Version 0.42.0"></a> <a href="https://modelcontextprotocol.io"><img src="https://img.shields.io/badge/MCP-Standard-orange.svg?style=flat-square" alt="MCP"></a> <a href="https://python.org"><img src="https://img.shields.io/badge/Python-3.10+-3776AB.svg?style=flat-square&logo=python&logoColor=white" alt="Python 3.10+"></a> <a href="https://pypi.org/project/agents-docs/"><img src="https://img.shields.io/pypi/v/agents-docs.svg?style=flat-square" alt="PyPI"></a> <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-green.svg?style=flat-square" alt="License"></a> </p>
Ultra-fast, zero-bloat local markdown documentation RAG for AI coding agents via MCP.
Instant BM25 search over official framework documentation and live AI model registries. Shared across Cursor, Claude Code, Antigravity, and Zed.
Why .agents/docs?
Coding agents frequently hallucinate deprecated APIs (Svelte 4 vs 5 Runes, Tailwind v3 vs v4, React 19 Actions) or reference outdated LLM models and token pricing.
Traditional vector-embedding RAG solutions are bloated: they require Docker containers, hundreds of megabytes of Python dependencies (Chroma, LangChain), and slow API embedding calls that split code blocks in half.
agents-docs applies the Lolaplex philosophy:
- Zero Vector DBs: Header-aware markdown chunking + pure Python lexical BM25 ranking with header match boost (+6.0) and full phrase boost (+10.0).
- Sub-10ms Search: Instant retrieval directly from plain markdown files on disk (
~/.agents/docs/). - Live AI Models Registry: Real-time tracked context windows, benchmarks, and token pricing ($/1M) for Frontier and Open-Weights models (Claude 5/4, GPT-5/o3/o4, Gemini 3.x, DeepSeek V4/R1, Qwen 2.5 Coder).
- Curated 21+ Framework Catalog: 1-click sync for official
llms.txt(TypeScript Handbook, Svelte 5, FastAPI, Tailwind v3/v4, Tauri 2, Next.js, Supabase, Hono, WXT, Netlify, Coolify, Cloudflare, Resend, React, Vite, Zod, Pydantic, shadcn/ui). - Deterministic Noise Pruning: Automatically strips HTML wrappers, link noise, and cookie banners without altering code blocks or tables.
- Universal MCP Server: FastMCP stdio interface with auto-sync and dependency manifest inspection.
Architecture & Flow
┌─────────────────────────────────────────────────────────────┐
│ DOCS & MODEL REGISTRIES │
│ Official llms.txt · TypeScript Handbook · OpenRouter │
└──────────────────────────────┬──────────────────────────────┘
│ agents-docs sync <name>
▼
┌─────────────────────────────────────────────────────────────┐
│ DETERMINISTIC NOISE PRUNER │
│ Strips HTML boilerplate while preserving 100% of code │
└──────────────────────────────┬──────────────────────────────┘
│ Header-aware chunking
▼
┌─────────────────────────────────────────────────────────────┐
│ LOCAL DOCUMENTATION STORE │
│ ~/.agents/docs/ai-models/ ~/.agents/docs/svelte-5/ │
│ ~/.agents/docs/typescript/ ~/.agents/docs/fastapi/ │
└──────────────┬───────────────────────────────┬──────────────┘
│ │
▼ ▼
┌─────────────────────────────┐ ┌─────────────────────────────┐
│ LEXICAL BM25 │ │ MCP SERVER & CLERK │
│ Header Match Boost (+6.0) │ │ search_docs · list_docsets │
│ Sub-10ms Exact Retrieval │ │ sync_project_docs │
└─────────────────────────────┘ └─────────────────────────────┘
Quickstart
1-Step Setup
pip install agents-docs && agents-docs init
Scaffolds ~/.agents/docs/, seeds the live ai-models registry, autowires MCP configurations into your installed IDEs, and registers assistant skills.
[!TIP] 🤖 Agent-Driven Setup (Zero Friction):
Simply tell your coding agent: "Install and set up agents-docs for me."
The agent installs the package, runsagents-docs init, and automatically retrieves fresh documentation whenever you ask technical stack or model pricing questions.
CLI Reference
| Command | Purpose |
|---|---|
agents-docs init |
Plug & Play setup: auto-configures MCP across all IDEs and seeds model docs |
agents-docs catalog |
Lists all 21+ pre-configured frameworks and model catalogs |
agents-docs sync <name> |
1-click download/update of official framework documentation |
agents-docs sync <name> --url <URL> |
Ingests any custom llms.txt, llms-full.txt, or markdown URL |
agents-docs search "<query>" |
Fast BM25 search across all installed docsets |
agents-docs search "<query>" --docset <name> |
Scoped search within a specific documentation set |
agents-docs prune |
In-place noise & boilerplate cleaner across all installed docs |
agents-docs list |
Lists all installed docsets on disk with file and byte counts |
agents-docs serve |
Runs the FastMCP stdio server (default) |
MCP Tools Reference
| Tool | Parameters | Description |
|---|---|---|
list_docsets |
None | Lists all installed docsets in ~/.agents/docs/ with file counts and byte sizes. |
search_docs |
query, docset (default: "all"), top_k (default: 4) |
Returns top matching markdown sections with headers and code blocks in milliseconds. Auto-syncs missing catalog docsets on demand. |
get_doc_page |
docset, rel_path |
Returns the raw markdown content of a specific file. |
list_catalog |
None | Lists all 21+ pre-configured frameworks and model catalogs available for 1-click sync. |
sync_docset |
name, url (optional) |
Downloads/updates official documentation from the catalog or custom llms.txt / markdown URL. |
sync_project_docs |
project_path |
Inspects package.json, requirements.txt, pyproject.toml, Cargo.toml and auto-syncs detected docsets. |
sync_memory_docs |
None | Scans ~/.agents/memory/PROJECTS.md and syncs all active stacks. |
Supported Ecosystem
- Claude Code: Bound via MCP server and
.agents/skills/docs-search. - Google Antigravity: Integrated via
.gemini/configrules andagents-docsMCP. - Cursor: Automatically configures
.cursor/mcp.jsonand agent rules. - Zed: Configures
context_serversand mirrors assistant skills. - VS Code / Windsurf: Autowires Cline / Roo-Code MCP configuration.
Open ABI Specification
Detailed architectural specifications live in abi/:
abi/WHY.md— Architectural rationale & BM25 vs Vector embeddings.abi/LAYOUT.md— Storage taxonomy in~/.agents/docs/.abi/MCP.md— Tool surface definitions and schemas.abi/CATALOG.md— Curated catalog schema and supported types.abi/LLMSTXT.md— Support for thellms.txtstandard.
Testing & Verification
Run the test suite across all engines:
python tests/run_all_tests.py
License
MIT License. See LICENSE for details.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.