MCP Overview

MCP is Itera's developer and agent integration surface for working with Itera through MCP clients and coding agents. Use it when an agent needs task context, planned implementation scope, prototype runtime details, or investigation history without building custom API glue.

Production server

The production MCP server is a remote service authenticated through OAuth:

https://api.iteradev.ai/mcp

Normal production setup must use the MCP client's OAuth connection flow. Do not ask users to paste user tokens, bearer tokens, custom Authorization headers, or SOCIALGRAPH_ACCESS_TOKEN into an MCP configuration.

See Connection setup for client configuration, OAuth discovery URLs, and a safe smoke test.

What the tools cover

The MCP surface is organized around the workflows agents need most:

• Iteration task context: read task summaries, timelines, blockers, prototype summaries, and implementation plans before deciding what to do.
• Iteration task creation: create a new task and queue its first product-phase run when starting new work from an MCP client is intentional.
• Planned PR execution: claim a dependency-ready planned PR and retrieve the scoped implementation context for that exact unit of work.
• Prototype debugging and control: inspect runtime logs, send prototype messages, start or fork prototype sessions, and recover active sessions when directed.
• Failure review: read bounded failure-review entries that summarize prior implementation attempts and relevant logs.

Safety

Some MCP tools mutate production state, including actions that create tasks, claim work, answer questions, review specifications, control prototypes, or send follow-up instructions. Start with read-only context tools whenever possible, confirm the target project, task, or prototype, and use mutation tools only when the intended state change is clear.

Read this section

• Start with Connection setup to connect a compatible MCP client through OAuth.
• Use the Agent guide for prompt templates, read-before-write workflows, and confirmation guidance.
• Use the Tool reference for available tools, required inputs, safety notes, and example prompts.
• Use Troubleshooting when discovery, OAuth, tool visibility, or structured tool errors need investigation.
• Use the agent-readable entry point when sharing compact MCP context with an agent.