InitRepo MCP Server

Last updated July 11, 2026

The InitRepo MCP Server exposes your generated project documents to any AI agent that supports the Model Context Protocol. Instead of manually copying planning documents into your agent's context at the start of each session, the MCP server makes them available as resources your agent can read on demand — structured, addressable, and always current.

What the MCP server does

MCP (Model Context Protocol) is an open standard that lets AI agents connect to external tools and data sources. The InitRepo MCP server implements this protocol so that AI agents — Claude Code, Cursor, Cline, and any other MCP-compatible tool — can read your project's planning documents directly, without you having to include them manually in each prompt.

The server exposes your InitRepo project as a set of structured resources: one per document (business analysis, PRD, architecture, etc.) plus a context index that maps the P#S# coordinate system across all documents. An agent can request the full architecture spec, or query the context index for everything relevant to a specific phase and step.

Available resources

Seven documents are exposed as readable resources your agent can fetch on demand:

  • Business Analysis — strategic context, problem definition, and market positioning
  • PRD — full product requirements with features, personas, and acceptance criteria
  • UX/UI Spec — design direction, component library, and UI block specifications
  • Technical Architecture — stack, components, data model, and API design
  • User Stories — all stories with acceptance criteria, organized by phase and step
  • Roadmap — phased implementation plan with sequencing and priorities
  • Context Index — the coordinate-addressed cross-document index for agent retrieval

Available tools

Beyond reading documents, agents can call four tools to navigate the build plan and track progress without loading the full document suite:

  • list_steps — returns the complete build plan (all phases and steps) with a done/not-done flag and indexed-context count per step. Use this first to orient on the plan before pulling any detail.
  • get_step_context(coordinate, full?) — retrieves the context anchored to a specific P#S# coordinate. Pass full=true to get the full markdown section from each source document rather than a short snippet — roughly 4% of the token cost of reading the whole suite.
  • search_spec(query, limit?) — keyword search across all deliverable documents, with each result tagged to its coordinate. Useful for finding the spec entry for a specific feature without knowing which document it lives in.
  • mark_step_complete(coordinate) — records that the agent has finished a build step. Drives the progress burndown visible on your Project page.

Connecting your agent

To connect an MCP-compatible agent to your InitRepo project, add the InitRepo MCP server to your agent's MCP configuration with your project ID and API key. The agent can then list and read your project's resources using standard MCP resource operations.

For Claude Code, add the server to your .claude/settings.json MCP servers configuration. For Cursor, add it to the MCP settings in Preferences. See MCP Authentication for credential setup.

Availability

API keys are created on any paid plan — generate one from your account page, then connect your agent using the authentication guide. Crucially, a key keeps working with the documents you've generated even after a subscription ends, so your AI agent never loses access to plans you already created.

Resources (document reads) and all four tools are available to any connected agent. The mark_step_complete tool writes build-progress state back to your project; all other operations are read-only.