LiteLLM Multi Model API
Route agent completions across 100+ models through LiteLLM from one MCP server instead of juggling separate provider SDKs.
Overview
LiteLLM MCP is an MCP server for the Build phase that gives agents unified access to 100+ LLMs through LiteLLM with optional proxy configuration.
What is this MCP server?
- LiteLLM Multi-Model API MCP server (v1.0.2) with Python/pypi stdio transport
- Documented access to 100+ LLMs including GPT-4, Claude, Gemini, and Mistral
- Supports direct provider keys (OPENAI_API_KEY, ANTHROPIC_API_KEY) plus optional LITELLM_API_BASE and LITELLM_API_KEY for
- BerriAI litellm-agent-mcp repository for agent-centric model switching
- Access to 100+ LLMs per server description
- Server package version 1.0.2 (litellm-mcp on pypi)
- Documented env vars: OPENAI_API_KEY, ANTHROPIC_API_KEY, LITELLM_API_BASE, LITELLM_API_KEY
What problem does it solve?
Indie agent builders waste time swapping provider SDKs and env keys every time they want to try GPT-4, Claude, Gemini, or Mistral in the same workflow.
Who is it for?
Solo builders shipping AI agents or copilots who want a single MCP surface for multi-vendor model routing and quick model experiments.
Skip if: Builders who only need one fixed model in Claude Code with no custom routing, or teams without any LLM API budget.
What do I get? / Deliverables
After registration, one stdio MCP server lets your agent call diverse models via LiteLLM using your configured API keys or LiteLLM proxy endpoint.
- Stdio MCP integration exposing LiteLLM-backed model calls to your agent
- Configurable path for direct vendor keys or centralized LiteLLM proxy usage
- Faster model swap experiments without per-provider client rewrites
Recommended MCP Servers
Journey fit
Canonical shelf is Build agent-tooling because the server exists to wire your product or coding agent to multi-model inference during implementation. Agent-tooling is the right facet for LiteLLM’s unified model API, optional proxy base URL, and provider key wiring over stdio MCP.
How it compares
Multi-model LLM gateway MCP server, not a prompt library skill or a hosted chat UI product.
Common Questions / FAQ
Who is LiteLLM MCP for?
It is for developers building agent features who want LiteLLM-backed access to many models from an MCP-compatible coding agent.
When should I use LiteLLM MCP?
Use it during Build and early Validate prototyping when you need to compare models or centralize inference through LiteLLM or a LiteLLM proxy.
How do I add LiteLLM MCP to my agent?
Install the litellm-mcp pypi stdio package, set the provider or proxy environment variables documented in server.json, and add the server to your MCP client configuration.