Lives in your repo.
Speaks to every coding agent.
CodeBroker connects over the Model Context Protocol, so whichever tool you code with, it already knows its way around your repo.
Three steps. Zero guesswork.
Your AI is great at writing code. It's bad at finding it. CodeBroker handles the finding.
It maps your repo
CodeBroker lives in your codebase and keeps a live map of every file, symbol and dependency — updated as your code changes.
Your AI asks it
Claude Code, Cursor or any MCP client asks questions like “where is auth handled?” or “what breaks if I change this?”
It returns the right context
Only the files and symbols that matter come back — a small, focused context window instead of the whole repository.
See it end to end.
From a cold repo to your AI answering with the right files — watch the whole loop, or click through it yourself.
The same question. Twice.
Watch Claude answer with nothing but grep, then with CodeBroker wired in. Same repo, same question — different amount of reading.
Less context. Fewer tokens.
Instead of dumping the repo into the context window, CodeBroker hands back only what's relevant — so your AI reads less and answers faster.
What your AI reads without it
36 files read · to answer one question
With no map of the repo, the search widens until something matches. Most of what gets opened is never used in the answer.