mle-kit-mcp

# MLE kit MCP [](https://pypi.org/project/mle-kit-mcp/) [](https://github.com/IlyaGusev/mle_kit_mcp/actions/workflows/python.yml) [](LICENSE) MCP server providing practical tools for ML engineering workflows, including local/remote bash, a text editor, file search, remote GPU helpers (via vast.ai), and an OpenRouter LLM proxy. ### Features - **bash**: Run commands in an isolated Docker container mounted to your `WORKSPACE_DIR`. - **text_editor**: View and edit files and directories in your workspace with undo support. - **glob / grep**: Fast filename globbing and ripgrep-based content search. - **remote_bash / remote_text_editor / remote_download**: Execute and edit on a remote GPU machine and sync files to/from it. - **llm_proxy_local / llm_proxy_remote**: Launch an OpenAI-compatible proxy backed by OpenRouter locally (in the bash container) or on the remote GPU. ## Requirements - Python 3.12+ - Docker daemon available (for `bash` tool) - ripgrep (`rg`) installed on the host (for `grep` tool) - `WORKSPACE_DIR` should be set with a path to working directory - Optional (for remote GPU tools): a `VAST_AI_KEY` with billing set up on vast.ai - Optional (for LLM proxy tools): an `OPENROUTER_API_KEY` ## Install Using uv (recommended): ```bash uv sync ``` Or standard pip install: ```bash python -m venv .venv && . .venv/bin/activate pip install -e . ``` ## Run the MCP server Set a workspace directory and start the server. The MCP endpoint is served at `/mcp`. ```bash WORKSPACE_DIR=/absolute/path/to/workdir uv run python -m mle_kit_mcp --port 5057 ``` Defaults: - `PORT` defaults to `5057` if `--port` is not provided - `mount_path=/` and `streamable_http_path=/mcp` ### Claude Desktop config ```json { "mcpServers": { "mle_kit": { …