MCP tools

Reference for the BugBrain MCP tool surface — the scope-gated tools an AI coding assistant can call, how a key's scopes decide which tools appear, and which tools are local-only.

The BugBrain MCP server connects AI coding tools (such as Claude Code, Cursor, and Codex) to your workspace. It exposes a scope-gated tool surface: the server reads your API key's scopes and registers only the tools those scopes permit. This page describes the representative tool categories — for setup, see MCP server.

How scopes decide the tool surface#

When the server starts, it inspects the API key's scopes and advertises only the tools that key can use. An assistant therefore never sees a tool it can't actually call. An empty scope list means full access, so every tool is registered. To change what's available, re-scope or re-mint the key (see API keys).

Tool categories#

The tools mirror the work you do in the app. Representative categories:

CategoryWhat the tools doRoughly maps to scopes
ProjectsList projects and read project context.Project read
Test runsStart exploratory runs and read run status and results.Runs read / write
IssuesList issues a run found and update their triage state.Issues read / write
Test casesList, author, and execute saved test cases.Test-case read / write / execute
Test plansList and run grouped sets of test cases.Test-plan read / execute
Accessibility & complianceStart scans and read audit results.Accessibility / compliance read / run
Local utilitiesLocal-only helpers, such as creating a tunnel to reach a local or staging app.Enabled via the local bridge

The exact list lives with the server

The precise tool names and inputs ship with the MCP package and the scopes you grant; this table is a representative overview. The tools your assistant actually sees are whatever your key's scopes advertise.

Local-only tools#

Some tools only make sense on the machine running the assistant and are enabled only through the local stdio bridge (npx @bugbrain/mcp), not the hosted transport. The clearest example is creating a tunnel so BugBrain's cloud runner can reach an app on localhost or a private staging URL. These local capabilities are turned on when the bridge runs in local mode and are never advertised by the remote, hosted server.

Frequently asked questions

How does the MCP server decide which tools I see?

The tool surface is scope-gated. The server reads your API key's scopes and registers only the tools those scopes permit, so an assistant is never offered a tool the key can't actually use. An empty scope list means full access, so every tool is advertised.

What kinds of tools are there?

Tools cover the same work you do in the app — listing and creating projects, starting runs and reading their results, listing and triaging issues, and managing test cases. The exact set depends on your key's scopes.

What are local-only tools?

Some tools only make sense on your own machine and are enabled solely through the local stdio bridge — for example creating a tunnel so BugBrain can reach a localhost or staging app. They are not advertised by the hosted transport.

How do I change which tools are available?

Re-scope or re-mint the API key the MCP server uses. Because tools are advertised from the key's scopes, narrowing or widening the scopes narrows or widens the visible tool set the next time the server starts.