Token Optimizer MCP
Stop agents repeating expensive calls and re-reading the same files
Compress tool outputs, logs and RAG chunks before the LLM sees them
A compression layer that sits between your coding agent and the model API, shrinking everything coming back from tools rather than the prompt you write. It targets developers running coding agents where tool output dominates token spend, delivering roughly 20% token savings for agents and 60-95% for JSON-heavy payloads with the same answers.
Headroom ships as a library, a proxy and an MCP server. In proxy mode you point the agent's base URL at it; each tool result, log file, source file or RAG chunk is passed through a compression algorithm before being forwarded to the model, and the compressed form is what re-enters the context window. In the video's test it cut a file to about 37% of its original size.
10 Repos conserve your token usage
Free CapCut, marketing tools, and all of github’s top 10 repos
Free clones: Notion, Lovable, Notebook LM + top 10 repos
Cut LLM cost by 95%, replace ElevenLabs, and 10 top GitHub repos
Stop agents repeating expensive calls and re-reading the same files
Move bulky tool output to disk instead of deleting it
Convert PDFs, Office files, audio and YouTube into clean Markdown
Stealthy adaptive Python scraping from one request to full crawl
Agent memory that learns instead of just recalling
Self-evolving context database giving agents organized, searchable memory