root_cause, steps, validation and risk_level are all fields on fixgraph_submit_fix, so what goes in is a structured account rather than a paste. The read side matches: search results carry a trust score alongside a fix count, so an issue with many fixes and a low score reads very differently from one with a single verified fix. Writing is the only thing gated — the API key is needed on submit, and neither search nor read requires one.
A FixGraph client with three tools — search issues with their trust scores and fix counts, read the verified fixes for one issue, and submit a new fix.
- fixgraph_search returns matching issues with trust scores and fix counts, filterable by category with a limit.
- fixgraph_get_fixes returns all verified fixes for an issue by its id or slug.
- fixgraph_submit_fix submits a fix with a title, root_cause, steps, validation and a risk_level.
Nothing for fixgraph_search or fixgraph_get_fixes. fixgraph_submit_fix requires a FIXGRAPH_API_KEY environment variable.
One command — npx -y fixgraph-mcp
