Install BugMojo in Claude Code
Install the plugin for nine skills, four subagents and three hooks, or add the hosted MCP server alone with one command. Either way, you sign in once in the browser.
Before you start
- Claude Code installed and working in your terminal.
- A BugMojo account you can sign in to in a browser.
- Access to at least one bug. The connection inherits your role, so Claude sees only the bugs you can already see.
Install the plugin
/plugin marketplace add viveksinra/bugmojo-sdk
/plugin install bugmojo@bugmojoRun both commands inside Claude Code.
When prompted, accept the BugMojo MCP URL default:
https://www.bugmojo.com/api/mcp?mode=qa.Run
/mcpand complete the BugMojo login in the browser that opens.Run
/bugmojo:setup. It proves the login, picks or creates the project, issues an API key for the machine paths, runsbugmojo initand wires the Playwright reporter, checking each step rather than assuming it worked.
MCP only, without the plugin
Open a terminal in the project you want Claude Code to work in and run the command below. It registers a server named bugmojo that talks to the hosted endpoint over HTTP.
Start (or restart) Claude Code so it picks up the new server.
Run
/mcpand complete the BugMojo login in the browser that opens. On the consent screen, choose the one workspace the agent may use and approve.Run
/mcpagain and confirm bugmojo is listed as connected.
claude mcp add --transport http bugmojo https://www.bugmojo.com/api/mcpThat gives Claude the tools without the skills or hooks. Add ?mode=qa for the regression, test-run and plan-review tools, ?mode=all for everything, and ?workspace=<slug> to pin one workspace. For first prompts, the full tool list, team setup and an API key for CI, read the guide How to Connect Claude Code to Your Bug Tracker via MCP.
What the plugin adds
| Skill | What it does |
|---|---|
/bugmojo:setup | Proves the login, picks the project, issues a key, runs bugmojo init, wires the reporter. |
/bugmojo:plan-tests | Interviews you on depth, target counts, routes, environments and who runs the manual cases, maps the codebase, proposes cases and pins them into a draft plan version. |
/bugmojo:generate-tests | Proposes cases for a scope, classifies them, and writes Playwright specs, verifying every selector against the running app first. |
/bugmojo:run | Checks the reporter is wired, runs the specs, confirms the run recorded, and triages each failure. |
/bugmojo:assign | Balances manual cases across the team, creates the run and writes each person's list. |
/bugmojo:report | Builds the client report: scope, what changed, coverage, results by mode, evidence, open bugs, and what was manual and why. |
/bugmojo:fix-bug <KEY> | Pulls the bug with its logs, gets an executable repro, localizes, fixes, opens a PR and links it. |
/bugmojo:triage-bugs | Ranks open bugs in the context of the plan: linked cases, defect density, reopens. |
bugmojo-conventions | Fires automatically when you touch tests/** or e2e/**: how to annotate a spec, and the flakiness rules. |
Four subagents keep the noisy work out of the main conversation: test-planner maps the codebase into proposals, test-generator writes and grounds specs, test-runner runs the suite and confirms it recorded, and evidence-reviewer triages each failure into BUG, FLAKY, ENVIRONMENT or TEST_DEFECT and files the real bugs.
The three hooks
- A session startsLikely cause: SessionStart hook.Prints your open bugs, assignments, recent failures and the agent inbox, so the work is in context before you ask.
- A Playwright run finishesLikely cause: PostToolUse hook.Auto-records the run when the reporter is configured, so you do not have to remember.
- A session tries to end after claiming tests passLikely cause: Stop hook: no run was recorded.The session is blocked with a reason. Record the run, or pass the override flag when you genuinely mean to stop.
Troubleshooting
- bugmojo does not appear in /mcpLikely cause: Claude Code has not reloaded its MCP config, or the entry is wrong.Restart Claude Code and run
/mcpagain. Still missing? Runclaude mcp get bugmojoto inspect the entry, re-run the add command and check the transport ishttp. - A tool you expect is missingLikely cause: Your role lacks its permission, or the mode switch is not set.Call
whoami(it shows your role and permissions) or reconnect with?mode=all. A tool is hidden when your role lacks its permission, which is not the same as the tool being unavailable. - The connection is refusedLikely cause: AI agents and MCP are turned off for the workspace.An owner or admin can turn AI agents & MCP back on under Settings, Security. While it is off, every MCP client is refused.
Common questions
Do I need an API key?
Not for the plugin itself. The hosted endpoint uses OAuth in the browser. You need a <code>bm_key_</code> key for the machine paths: the Playwright reporter and CI.
What is the difference between this and just adding the MCP server?
The MCP server gives Claude the tools. The plugin adds the skills, the subagents and the hooks: the workflow around those tools.
Why does my session refuse to finish?
The Stop hook blocks a session that ran Playwright or claimed tests pass without recording a run. There is an explicit override flag when you really mean it.
Can I run the plugin from a local checkout?
Yes: <code>claude --plugin-dir ./packages/claude-plugin</code> loads it straight from the monorepo.
Do I need to install a separate package for MCP only?
No. BugMojo hosts the MCP server at <code>https://www.bugmojo.com/api/mcp</code>, so <code>claude mcp add</code> points Claude Code straight at it and nothing is installed locally.
The server does not show up after I add it. What now?
Restart Claude Code so it reloads its MCP config, then run <code>/mcp</code> again. If <code>bugmojo</code> is still missing, run <code>claude mcp get bugmojo</code> to inspect the entry, re-run the add command and check the transport is <code>http</code>.
Related articles
- Install BugMojo in CursorOne click from the install deeplink, or a four-line mcp.json, plus a rule that only loads when you are in test code.
- MCP tool referenceKnow exactly which tools a connected AI agent can call over MCP to read a BugMojo bug and act on it.
- Agent access, tokens, and the security modelUnderstand how an AI agent's access is scoped, what it inherits from you, and the guardrails that keep it in bounds.
- The agent inbox and task leasesA shared, risk-ranked queue any agent can read, and an exclusive lease so two sessions never work the same task.
- Record Playwright runs with the reporterThree lines in playwright.config.ts and two environment variables turn every test run you already do into recorded, provable results.