Connect Zentrik MCP

Use Zentrik MCP when you want your assistant to work from shared product context, not just the contents of the current chat.

For direct MCP clients, use:

• Server URL: https://zentrik.ai/mcp
• Transport: Streamable HTTP when the client asks
• Auth: browser-based OAuth
• Scope: one connection = one workspace
• Open MCP Registry: listed as ai.zentrik/mcp

Use a published listing or package for a client-specific setup flow:

• ChatGPT and Codex: Open the Zentrik listing in the OpenAI Plugins Directory. The listing uses a reviewed OpenAI app snapshot.
• Claude: Open Zentrik in the Claude Connectors Directory. The connector uses the generic server URL above.
• Claude Code: Install the Zentrik Claude Code plugin from its public marketplace.
• Codex CLI and other Agent Plugins clients: Use the portable Zentrik Agent Plugin.

For any other MCP client, use the server URL and OAuth flow above. A live endpoint change does not automatically update the reviewed ChatGPT snapshot.

Start with the guide for the tool you are using:

• Codex setup guide
• Cursor setup guide
• Claude setup guide
• ChatGPT setup guide

Quick fit:

• Codex or Codex CLI if you want the shortest hosted setup flow or the portable Agent Plugin
• Cursor if you want Zentrik context available inside your editor and agent workflows
• Claude or Claude Code if you want product context while researching, synthesizing, drafting, or building
• ChatGPT if you want the reviewed Zentrik app in the shared Plugins Directory

If you are in another MCP client, use the same server URL and OAuth flow. If the client only asks for a static bearer token or API key, it is not using the right auth flow for Zentrik MCP.

Start with small, evidence-seeking requests:

• ask what Zentrik already knows about a problem, request, or idea
• pull the relevant opportunity, idea, or context record before you plan or build
• add fresh notes from a call, interview, or demo without leaving your current tool
• run a Study learning loop: find or create a private plan, revise it, monitor de-identified evidence, and generate draft findings while keeping launch and participant distribution in Zentrik
• review and answer Idea planning questions, maintain Idea taxonomy and Initiative links, and regenerate generated Initiative documents after approved source changes
• ask what a person or agent changed, with confirmed, no-change, or unverifiable outcomes and explicit coverage

Keep in mind:

• access is tied to the workspace you approved
• your role in that workspace still matters
• access to initiatives and linked Studies follows your available teams; the agent can use teams.list to resolve safe team IDs before creating or filtering work
• Workspace Owners and Admins can replace an initiative’s complete team assignment with initiatives.set_teams; reconnect if an older OAuth grant does not include newly released tools
• if you want a different workspace, reconnect and choose it
• if newly released tools do not appear, reconnect and approve the current permissions; some clients keep the original grant until you do
• workspace API keys still work for /external/v1, but not for /mcp
• every MCP write creates a durable receipt; the same shared boundary covers product writes reused by Zentrik agents
• use system.get_change_history for product changes and system.get_activity for background jobs
• change history states whether the requested time and channels are complete, partial, or untracked and summarizes repeated attempts so retry noise does not hide other outcomes
• MCP and Slack-agent product mutations are fully tracked after receipt collection begins; workspace-agent history is partial; UI, External API, integration, background-job, and portal mutations are untracked

Next:

• Product management MCP servers compared for a dated, source-linked view of Zentrik, Productboard, Aha!, Linear, Atlassian, PostHog, and LaunchDarkly
• Product-to-spec use case for how signal becomes build-ready scope
• Codex handoff use case for implementation-ready context
• Claude Code handoff use case for clearer requirements prompts
• MCP use cases for practical first prompts
• MCP best practices for lighter prompting and realistic limits