AI agent
Codex + ProjectMind
Chained agent workflows
Codex orchestrates multi-step delivery; ProjectMind is the system of record for tasks, epics, and agent jobs.
Example chat
via ProjectMind MCP · Codex
How to connect Codex
- In Codex, add a custom MCP connector pointing at the ProjectMind server URL (see below).
- Sign in with ProjectMind OAuth or supply a project API key in the Authorization header when OAuth is unavailable.
- Confirm tools/list includes ProjectMind tool names (listTasks, getProjectContext, enqueueAgentJob, …).
- Scope to one project with ?projectSlug=your-slug or set a default project in Settings → Preferences.
- Ask the agent to call getProjectContext or listTasks before createTask, startEpicExecution, or enqueueAgentJob.
ProjectMind MCP server URL
https://projectm.dev/api/mcp
Claude Code CLI
claude mcp add --transport http imemory-api https://projectm.dev/api/mcp --header "Authorization: Bearer imk_..."
OAuth sign-in is recommended for chat clients. Project API keys (imk_…) work when the client cannot open a browser. Confirm before startEpicExecution, enqueueAgentJob, or deleteTask.
What you can do with Codex
Task inbox
listTasks and getCurrentTask to see what is active before changing status.
Context search
searchContext and getTaskContext to ground answers in your project memory.
Epic planning
listEpics, getEpic, and confirmEpicPlan to break work into tracked stages.
Agent jobs
resolveAgentProfile and enqueueAgentJob to queue coding agents—with your confirmation.
Example prompts
- List listAgentJobs filtered to running and summarize phases.
- Refresh getVercelRuntimeErrors and createImprovementProposal if errors spiked.
Prefer read tools first (listTasks, getProjectContext, searchContext). Mutations should follow your confirmation—especially epic execution and agent job enqueue.
Frequently asked questions
- Is ProjectMind preinstalled in ChatGPT or Claude?
- No. Add ProjectMind as a custom MCP connector (or sign in with OAuth when the client supports remote MCP). Use your ProjectMind account and approve project access in the consent flow.
- Will the agent change my project without asking?
- Destructive or high-impact MCP writes—deleting tasks, starting epic execution, completing agent jobs—should run only after you confirm in chat or in the ProjectMind UI. Prefer read tools first (listTasks, getTaskContext, searchContext).
- Which authentication should I use?
- OAuth is best for interactive assistants. Project API keys (imk_…), global keys (imgk_…), and bucket keys (imbk_…) work in headers for headless clients. Never put keys in query strings.
- Can agents run epics and background jobs?
- Yes, when your role allows it. Tools such as confirmEpicPlan, startEpicExecution, and enqueueAgentJob enqueue work your team can monitor in ProjectMind. Pause or cancel from the app if something looks wrong.