Vitex2d ago

Udělátko co se tváří jako #MCP server a zpřístupňuje tak (lokálně běžící) #Ollama

Zejtra až se mi obnoví kredity tak to vyzkouším. Zázraky od toho nečekám ale kdyby to zastalo alespoň něco práce tak by to bylo fajn.

https://github.com/Vitexus/cc_token_saver_mcp

GitHub - Vitexus/cc_token_saver_mcp: Allow Claude code to use local llm for smaller tasks to save token or for specialized task

Allow Claude code to use local llm for smaller tasks to save token or for specialized task - Vitexus/cc_token_saver_mcp

GitHub
Show threadVitex

❯ use #cc-token-saver mcp to summarize commit message

● All 9 tools are now live. Let me load the schemas and grab the recent commits to summarize.

Called cc-token-saver, ran 1 shell command

● It works — qwen2.5-coder:7b via Ollama responded directly through the MCP tool. The local LLM summarized the 10 commits in one call, no Claude tokens spent on it.

Now let me also try generate_commit_message on the actual last diff so you can see it end-to-end:

Called cc-token-saver, ran 2 shell commands

● Both tools working perfectly via Ollama. Here's what just happened:

- summarize_text — condensed 10 commit log lines into a 60-word paragraph, focusing on "what changed and why"
- generate_commit_message — read the diff and produced refactor(debian): move and update files for package structure with a two-line body, following Conventional Commits — entirely on your local GPU, zero Claude tokens spent

#ItWorks #Agent #LLM #MCP

Jun 26 at 4:53amWeb