How to reduce Claude Code token usage

Before an agent edits anything it has to learn your codebase: list folders, search for names, open files, then open the files those import. On a large repository that exploration can be most of the task. Every file it reads stays in the conversation and is sent again with each later step, so an early wrong turn is paid for many times.

These work with or without any tool:

• Name the files: “fix the refresh check in auth/session.service.ts” beats “fix login”.
• Start fresh between tasks (/clear), so old context isn't re-sent.
• Ask for a plan first, then the edit: a plan is cheap, a wrong edit and its fix are not.
• Keep CLAUDE.md short: it is loaded on every turn.
• Say which tests to run instead of letting it run the whole suite.
• Split big tasks: one change per conversation.

Synapse keeps a map of your repository (files, functions, who calls what, which tests cover what) and the history behind it (pull requests, reviews, commits). Over MCP, Claude Code gets a short brief before it starts: the files and line ranges to change, the callers, the pattern to follow and the tests to run. The exploring step mostly disappears.

In our benchmark (Claude Code with Sonnet, 7 real tasks on a repository Synapse had never seen, 2 runs each), Synapse cut tokens by 58% and cost by 48%, and every edit still compiled. Planning tasks saved the most. Results vary by repository and task.

Synapse is free to start: connect a GitHub repository and add it to Claude Code, Cursor or Claude in about two minutes. Create an account or read the docs.