mcp-server-cc-token-saver¶

An MCP server that delegates simple tasks to a local LLM (any OpenAI-compatible endpoint, for example Ollama or LM Studio) so that a cloud assistant spends fewer tokens. Companion package: mcprack-mcp-server-cc-token-saver. Binary: /usr/bin/mcp-server-cc-token-saver.

sudo apt install mcprack-mcp-server-cc-token-saver

Configuration¶

Variable	Required	Meaning
`OPENAI_BASE_URL`	no	OpenAI-compatible API base URL (default `http://localhost:1234/v1`)
`OPENAI_API_KEY`	no	Key sent to the endpoint (default `none`; usually ignored; secret store)
`LOCAL_MODEL_NAME`	no	Model to request (default `qwen2.5-7b-instruct`)
`LOCAL_LLM_TEMPERATURE`	no	Sampling temperature (default `0.7`)
`LOCAL_LLM_MAX_TOKENS`	no	Maximum tokens per response (default `-1`, no limit)

Registration:

sudo mcprack server show mcp-server-cc-token-saver
sudo mcprack server edit mcp-server-cc-token-saver --add-secret-key <SECRET_VARIABLE>
sudo mcprack secret set mcp-server-cc-token-saver <SECRET_VARIABLE>

Replace <SECRET_VARIABLE> with each secret variable from the table.

Read-only behaviour: All tools only query the local model; none changes external state.

Tools¶

query_local_llm, query_local_llm_with_context, list_available_models, switch_model, summarize_text, generate_commit_message, generate_unit_tests, explain_code, translate_text.