# LibreChat MCP > LibreChat MCP connects your self-hosted AI instances to any agent client. It lets you manage custom agents, list all available models, and generate chat completions using an OpenAI-compatible interface for your private LLM setups. ## Overview - **Category:** productivity - **Price:** Free - **Tags:** llm-orchestration, chat-interface, self-hosted, model-management, ai-agents, unified-api ## Description Connecting your own LibreChat environment through this MCP gives your AI client direct control over your private language model ecosystem. Instead of relying on a single provider's features, you can treat your self-hosted agents like any other tool right inside Claude or Cursor. You first use the `login` tool to authenticate with your credentials and grab necessary access tokens. Once authenticated, you can list everything available using `list_models`. After that, you send requests via `chat_completions` to run chats against specific agents. Need structured data? The `open_responses` tool handles that too. This makes building complex agent workflows simple, letting your AI client talk directly to the models you set up yourself. It’s a solid way to centralize access to multiple private LLMs through one secure interface on Vinkius. ## Tools ### chat_completions Creates a chat completion using the Agents API, allowing you to send prompts and get model responses. ### list_models Retrieves a list of all available models and agents configured in your LibreChat setup. ### login Authenticates you with your credentials, retrieving necessary access and refresh tokens for the session. ### open_responses Generates a structured AI output using the Open Responses API specification. ## Prompt Examples **Prompt:** ``` List all available agents in my LibreChat instance. ``` **Response:** ``` I've retrieved the available agents. You have access to: 'Research-Assistant' (ID: agent_123), 'Code-Interpreter' (ID: agent_456), and 'General-Chat' (ID: agent_789). ``` **Prompt:** ``` Login to LibreChat using my credentials. ``` **Response:** ``` Authentication successful. I have obtained the access and refresh tokens for your session. You can now proceed with querying models or creating completions. ``` **Prompt:** ``` Ask agent_123 to summarize the latest trends in AI. ``` **Response:** ``` Sending request to agent_123... The agent responded: 'The current trends in AI include the rise of multi-modal models, autonomous agentic workflows, and increased focus on efficient small language models (SLMs)...' ``` ## Capabilities ### List Available Models You can ask the MCP for a full list of every agent and model configured in your LibreChat instance. ### Run Chat Completions The MCP sends messages to specific agents, generating responses that mimic standard chat completion APIs. ### Generate Structured Responses You prompt the agent for output and receive a highly structured response using the Open Responses API format. ### Manage Authentication The MCP allows you to log in directly with your email and password, obtaining required access tokens without needing static API keys. ## Use Cases ### Building an internal research assistant A researcher needs their agent client to talk to five different specialized models. Instead of configuring five separate API endpoints, they connect the LibreChat MCP, run `list_models` to verify all agents are online, and then use `chat_completions` to route queries dynamically based on the prompt topic. ### Automating compliance reporting A DevOps team needs an agent to summarize operational data into a strict JSON format. They connect via LibreChat MCP and use `open_responses`, forcing the model output into a predictable, machine-readable structure that can be automatically ingested by another system. ### Integrating legacy LLMs A developer has an older, specialized agent running on custom hardware. They connect it through LibreChat MCP and expose its functionality via `chat_completions`, allowing modern tools like Cursor to interact with the old system without code changes. ### Testing new agents quickly A power user wants to test an experimental agent before deploying it widely. They first use `login` for quick authentication, then run `list_models` to find the specific ID, and finally execute a few test prompts using `chat_completions`. ## Benefits - You get centralized control over multiple private LLMs. Instead of managing separate API keys for every model, you connect once to the LibreChat MCP and gain access to everything. - The `chat_completions` tool lets you run complex chats against agents using a standard OpenAI-compatible interface, making integration painless for your developers. - Authentication is easier with `login`. You can log in directly using credentials (email/password) instead of relying solely on static API keys. - Structured data output is guaranteed via the `open_responses` tool. If you need JSON or a specific schema from an agent, this ensures your response format is consistent every time. - The `list_models` capability instantly shows you what agents are available, saving you the headache of guessing which IDs to use in your workflows. ## How It Works The bottom line is, it gives your AI client programmatic access to run and manage every model in your self-hosted LibreChat environment. 1. Subscribe to the LibreChat MCP and provide your Base URL. If you don't have an API key, use the `login` tool with your email and password. 2. Use `list_models` to confirm that all desired agents are visible to your AI client. 3. Send a prompt using either `chat_completions` or `open_responses`, targeting the specific agent ID you need. ## Frequently Asked Questions **How do I get started with the LibreChat MCP?** You start by subscribing to the MCP and running the `login` tool using your credentials. This authenticates you and gives your agent client the necessary tokens for subsequent calls. **Can I use chat_completions with agents that aren't OpenAI-compatible?** Yes, that is the point of this MCP. It provides an OpenAI-compatible interface layer over your self-hosted LibreChat environment, making diverse models appear uniform to your client. **Is there a way to check which agents are active in my LibreChat instance?** Use the `list_models` tool. It provides an immediate overview of all available agents and model configurations, saving you from guessing IDs. **What if I need the output data to be perfectly formatted JSON?** For guaranteed structure, use the `open_responses` tool. This API is specifically designed to force your agent's response into a predictable format that downstream systems can read. **Does this MCP require me to modify my core AI client code?** No. You connect the LibreChat MCP through your existing compatible client (Claude, Cursor, etc.). The tool handles the complex routing and API translation for you.