ProductBrain

What ProductBrain MCP server does

ProductBrain MCP server gives MCP clients access to a ProductBrain product plan. It covers structured planning data such as goals, needs, bets, and jobs, along with phases and project membership. The server exposes read, write, workflow, administration, and sharing operations through MCP tools.

For plan discovery, search provides semantic lookup and is intended as the first tool for searches. list_nodes supports bulk reads with optional type and iteration filters, while get_tree returns a node with surrounding context such as ancestors, siblings, children, or a subtree. Changes go through mutate, which can create, update, or delete nodes and phases.

This fits the Product Management category when an agent needs to maintain a shared product plan rather than only answer questions about static documentation.

How it works

The ProductBrain MCP server is a thin transport layer over ProductBrain's versioned v1 REST API. Each MCP tool corresponds to an API endpoint, and the API remains the underlying contract. Responses pass through the API's _meta coaching data so clients receive the same guidance returned by the REST service.

The plan tools include iteration lookup, story-mapping workflows, OKF export, and changelog access. set_view and read_view control and inspect the live canvas. Project tools list, create, rename, and archive projects; archiving is reversible. Webhook tools manage registrations and secrets, while share-link tools create public read-only links, audit them, and revoke them.

Membership tools list project members, add agent seats, invite contributors, and revoke access. tier_status reports the account tier, API access, credit balance, and limits. Team-only operations return a tier error when used with a Builder key.

Setup and configuration

Install the published npm package with an MCP client's standard server configuration. The README provides examples for Claude Desktop and Claude Code using npx -y @productbrain-com/mcp; Cursor and other clients use the same command, arguments, and environment entries.

Set PRODUCTBRAIN_API_KEY to a pb_ key from the ProductBrain application. PRODUCTBRAIN_PROJECT_ID is optional because tools can receive a project ID directly. PRODUCTBRAIN_API_URL is also optional and defaults to https://productbrain.com/api/v1.

The repository includes headless spike and parity scripts for testing against a real deployment. The documented approach uses a throwaway project, and the parity script archives its temporary project because no project-delete endpoint exists.

Tools and capabilities

• Search and inspect plan data with semantic search, node lists, and contextual trees.
• Create, update, and delete nodes or phases, with idempotency support for mutate retries.
• Run the Story-Mapping workflow using add, task-curate, or phase-assign operations.
• Export the plan as an OKF file bundle and inspect its changelog.
• Manage projects, the live canvas, webhooks, public read-only links, members, agent seats, and contributor invitations.

Limitations and notes

create_project is not retry-safe because its project-creation operation reads the idempotency header after the relevant mutation step. Supplying an explicit project ID is recommended. Webhook creation returns the signing secret only once, and adding an agent seat also returns its pb_ key only once.

Public share links are read-only but accessible to anyone with the link, so revoke links that are leaked or no longer needed. Agent access is removed when the agent's last project membership is revoked. Team-tier tools require the Team tier. The source repository is MIT licensed, and the MCP registry name is io.github.moxzas/productbrain.