
Lm
Hand routine coding work from Claude to a local or cheaper OpenAI-compatible model
- Pricing
- Open source
- Platforms
- CLI
- Runs
- Locally
- Version
- 3.3.4
- Listing updated
- 1 day ago
About Lm
Lm is an open-source MCP server for ai models and llm tooling. It runs locally on your machine and is started by your AI client.
Houtini LM exists to cut token spend: Claude keeps the planning and judgement, while bounded jobs such as boilerplate, code review, commit messages and format conversion go to another model and come back for Claude to check. The target can be a local model on your GPU, OpenAI GPT models, a LiteLLM router, OpenRouter or any OpenAI-compatible endpoint, set through environment variables for URL, key, model and provider.
With a tool such as code_task_files, source files are read by the houtini-lm process and sent straight to the other model, so they never enter Claude's context. The author's benchmark on TypeScript files shows reviews using roughly 95% fewer tokens in Claude's window, 93.3% averaged across a session, with the caveat that small tasks save less.
Version 3.3 adds direct OpenAI support (sending the parameters those reasoning models accept), reads the real model behind LiteLLM aliases, and lets you set thinking to auto, off or on. It is Apache-2.0 licensed and ships a Docker guide and manual.
Install
Run in a terminal
claude mcp add lm --env HOUTINI_LM_API_KEY='<HOUTINI_LM_API_KEY>' -- npx -y @houtini/lmAdd to claude_desktop_config.json
{
"mcpServers": {
"lm": {
"command": "npx",
"args": [
"-y",
"@houtini/lm"
],
"env": {
"HOUTINI_LM_API_KEY": "<HOUTINI_LM_API_KEY>"
}
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"lm": {
"command": "npx",
"args": [
"-y",
"@houtini/lm"
],
"env": {
"HOUTINI_LM_API_KEY": "<HOUTINI_LM_API_KEY>"
}
}
}
}Add to .vscode/mcp.json
{
"servers": {
"lm": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@houtini/lm"
],
"env": {
"HOUTINI_LM_API_KEY": "<HOUTINI_LM_API_KEY>"
}
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"lm": {
"command": "npx",
"args": [
"-y",
"@houtini/lm"
],
"env": {
"HOUTINI_LM_API_KEY": "<HOUTINI_LM_API_KEY>"
}
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"lm": {
"command": "npx",
"args": [
"-y",
"@houtini/lm"
],
"env": {
"HOUTINI_LM_API_KEY": "<HOUTINI_LM_API_KEY>"
}
}
}
}Configuration
HOUTINI_LM_ENDPOINT_URLBase URL of the OpenAI-compatible API endpoint (e.g. http://localhost:1234, https://openrouter.ai/api). Legacy alias: LM_STUDIO_URL.HOUTINI_LM_API_KEYBearer token for API authentication (no auth if blank). Legacy aliases: LM_STUDIO_PASSWORD, LM_PASSWORD, OPENROUTER_API_KEY.HOUTINI_LM_MODELModel identifier to use for requests (auto-detected if not set). Legacy alias: LM_STUDIO_MODEL.HOUTINI_LM_PROVIDEROptional provider override: 'openrouter' forces OpenRouter-specific handling (attribution headers, reasoning.exclude, no inference serialisation). Auto-detected from the endpoint URL when possible.
Tools this server provides (1)
Taken from the project documentation; the running server may differ.
- code_task_files
- Delegate a review or task on source files to another model without loading them into Claude's context
What it does
- Model offloading
- Delegates drafting and review to a model you choose.
- Wide endpoint support
- Works with local servers, OpenAI, LiteLLM and OpenRouter.
- Files kept out of context
- Delegated source files bypass Claude's context window.
- Model-aware requests
- Adjusts parameters for OpenAI reasoning models and router aliases.
- Thinking control
- Set reasoning to auto, off or on.
- Second-opinion reviews
- Gets another model to review your code on demand.
Good for
- Offload commit messages and boilerplate
- Get a second model to review a diff
- Cut token cost on overnight refactors
- Use a local GPU model for drafts
- Route work through a LiteLLM fleet
Pricing
Open source
Free npm package; you pay your chosen model provider
Prices change. Check houtini.com before you buy.
Lm FAQ
What endpoints work?
Any OpenAI-compatible API, including LM Studio, OpenAI, LiteLLM and OpenRouter.
How is it configured?
Environment variables set the endpoint URL, API key, model and provider.
What licence is it?
Apache-2.0.
How much does it save?
The README's benchmark shows about 93% fewer tokens in Claude's context on delegated reviews.
Is Lm free?
Yes. Lm is open source, so you can use and self-host it for free.
How do I add Lm to Claude Code?
Run this command in your terminal: claude mcp add lm --env HOUTINI_LM_API_KEY='<HOUTINI_LM_API_KEY>' -- npx -y @houtini/lm
What are the best alternatives to Lm?
The closest alternatives to Lm are REA, Codebase Memory, GitHub, Serena MCP: the IDE for your agent and XERJ.
Alternatives to Lm

REA
MCP serverCLI and MCP server that lets coding agents reverse engineer apps and binaries
Runs locally- Open source
- 464K GitHub stars
- Developer tools

Codebase Memory
MCP serverLocal MCP server that builds a code knowledge graph for AI coding agents
Runs locally- Open source
- 48.7K GitHub stars
- Knowledge and memory

GitHub
MCP serverOfficial MCP server connecting AI assistants to GitHub repos, issues and workflows
Not checked yet- Remote
- Open source
- 34.5K GitHub stars
- Developer tools

Serena MCP: the IDE for your agent
MCP serverMCP toolkit giving coding agents symbol-level code retrieval and editing
Runs locally- Open source
- 31.5K GitHub stars
- Developer tools

XERJ
MCP serverLocal search engine for AI agents with cited answers over your code and docs
Runs locally- Open source
- 19.2K GitHub stars
- Knowledge and memory

Ouroboros
MCP serverSpec-first layer that pins requirements and verifies results of AI coding agent work
Runs locally- Open source
- 6.4K GitHub stars
- Developer tools