Itera MCP Agent Entry Point

Use Itera MCP when an agent needs to create Itera tasks, read task context, inspect implementation-plan scope, debug prototypes, or review failure history through a compatible MCP client.

Server and authentication

• Production MCP endpoint: https://api.iteradev.ai/mcp
• Production authentication: OAuth only, using the MCP client's remote-server OAuth flow.
• Do not configure manual user tokens, bearer tokens, generic SocialGraph API tokens, custom Authorization headers, SOCIALGRAPH_ACCESS_TOKEN, or token environment variables for normal production MCP setup.

Start with the MCP overview, then use connection setup for OAuth discovery URLs, client setup, and the safe smoke test.

Core references

• MCP overview: production endpoint, workflow coverage, and safety model.
• Connection setup: URL-only remote MCP setup, OAuth flow, and smoke test.
• Tool reference: exact tool names, inputs, classifications, and safety notes.
• Agent guide: read-before-write workflow and prompt templates.
• Troubleshooting: discovery, OAuth, tool visibility, and structured tool errors.

Safety rules for agents

• Read before write. Search, summarize, inspect blockers, inspect the plan, read timelines, list prototypes, or review logs before changing state.
• Confirm explicit targets before mutations. State the project ID for task creation, or the canonical task ID, planned PR ID, specification ID, prototype ID, session ID, operation, and expected effect for existing work.
• Treat creating tasks, planned PR claiming, answering questions, reviewing specifications, submitting follow-ups, starting prototypes, sending prototype messages, controlling prototype sessions, and marking planned PRs merged as production state changes.
• For connection smoke tests, use deliberately nonexistent IDs only. The documented safe smoke test should return a not-found error without claiming real work.
• Do not use real READY_UNCLAIMED planned PR IDs for testing unless the user intends to claim that exact planned PR for implementation.

Tool selection guidance

• To find work, use task search first, then summarize the selected task.
• To start new work, confirm the project ID and initial intent before creating an iteration task.
• To understand why work is blocked, inspect blockers and the timeline before answering questions or reviewing specifications.
• To prepare implementation, read the task plan and preview planned PR context before claiming a planned PR.
• To debug prototypes, list prototypes and read logs before starting, messaging, stopping, retrying, rebasing, or checking out a commit.
• To investigate repeated failures, read failure-review entries and compare them with the current task plan before taking action.
• When a tool call fails, use the structured status code as the durable signal. After a failed mutation attempt, read current state before retrying.