# llmmcp [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/zero-abd/llmmcp  
**GitHub Stars:** 6  
**npm Downloads (last month):** 39  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/llmmcp

## Description
MCP server for real-time LLM API documentation — stop hallucinations

## Tools
Capabilities this server exposes over MCP:

- **search_docs** — Search real-time LLM API documentation to get accurate, up-to-date information. Use this to verify API parameters, model names, pricing, and capabilities for Gemini, Claude, and OpenAI instead of relying on training data.
- **list_llms** — List available LLM documentation providers and their latest known model versions. Contains data for OpenAI, Anthropic, and Google Gemini.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "llmmcp": {
    "command": "npx",
    "args": ["-y","llmmcp@latest"]
  }
}
```

## Documentation

## What llmmcp MCP server does

The llmmcp MCP server exposes current documentation for three major LLM providers: OpenAI, Anthropic, and Google Gemini. It is intended for AI coding clients that need dependable answers about model availability, API behavior, tool-use syntax, context limits, rate limits, and current implementation patterns.

The service addresses a common documentation problem: an agent may produce code based on deprecated libraries, unavailable models, or older API conventions. Rather than embedding a fixed provider list in the client, llmmcp retrieves provider and model information dynamically.

## How it works

An indexing process runs weekly and collects Markdown or text from official provider documentation. The collected material is divided into chunks, embedded, and stored in Pinecone. Pinecone's integrated embedding support is used for the vector database layer.

When an MCP client sends a search request, a Cloudflare Worker handles query embedding and document retrieval. Frequently requested results can be cached in Cloudflare Workers KV. The command-line client acts as a thin MCP layer that forwards requests to the worker's API.

This architecture means search results come from an indexed documentation corpus rather than from the language model's built-in knowledge alone. The index is refreshed weekly, so the service is designed for current reference information but is not described as an instant mirror of provider documentation.

## Setup and configuration

The llmmcp MCP server can be launched through `npx`, without a separate local installation. For Cursor, add an MCP server named `llmmcp` in the MCP server settings and use `npx -y llmmcp@latest` as the command.

Claude Desktop can use the same package through an MCP configuration entry with `npx` as the command and `-y`, `llmmcp@latest` as its arguments. The README also identifies Windsurf as a supported agent client, although it does not provide a separate Windsurf configuration example.

The provided setup instructions do not require environment variables or API keys for client use. Self-hosting the backend is mentioned as a contribution option, with deployment details referred to in the repository documentation.

## Tools and capabilities

- `search_docs` searches the latest indexed official documentation for specific technical questions, such as tool-use parameters for a particular Gemini model.
- `list_providers` returns a dynamically updated provider list and the models currently promoted by those providers.

These tools make the llmmcp MCP server suitable for agents that generate or review integrations with OpenAI, Anthropic, or Google Gemini. The available sources are limited to the providers and documentation indexed by the project.

## Limitations and notes

It does not document arbitrary web search, access to private provider documentation, or direct execution of provider API calls. The documentation corpus is refreshed weekly, so changes made after the latest indexing run may not be present immediately.

The project is open source and licensed under MIT.

_Full upstream README: https://allmcps.com/mcp/llmmcp/readme_