---
title: "Connect Zentrik MCP"
canonical_url: https://zentrik.ai/docs/integrations/mcp
markdown_url: https://zentrik.ai/docs/integrations/mcp.md
category: "Integrations"
learning_track: null
last_reviewed: 2026-07-06
---

# Connect Zentrik MCP

Connect your MCP-capable tool to the same product context your team already keeps in Zentrik.

## Overview

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

The shared values are the same everywhere:

- **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`
- **Claude:** use the Connectors Directory when available, or ask a Team/Enterprise admin to add Zentrik as an organization custom connector

## Choose your client

Start with the guide for the tool you are using:

- [Codex setup guide](https://zentrik.ai/docs/integrations/mcp-codex)
- [Cursor setup guide](https://zentrik.ai/docs/integrations/mcp-cursor)
- [Claude setup guide](https://zentrik.ai/docs/integrations/mcp-claude)

Quick fit:

- **Codex** if you want the shortest hosted setup flow
- **Cursor** if you want Zentrik context available inside your editor and agent workflows
- **Claude** if you want product context while researching, synthesizing, or drafting

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.

## Once connected

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

Keep in mind:

- access is tied to the **workspace** you approved
- your role in that workspace still matters
- if you want a different workspace, reconnect and choose it
- workspace API keys still work for `/external/v1`, but not for `/mcp`

Next:

- [Product-to-spec use case](https://zentrik.ai/use-cases/customer-feedback-to-specs) for how signal becomes build-ready scope
- [Codex handoff use case](https://zentrik.ai/use-cases/codex-product-context) for implementation-ready context
- [Claude Code handoff use case](https://zentrik.ai/use-cases/claude-code-product-context) for clearer requirements prompts
- [MCP use cases](https://zentrik.ai/docs/integrations/mcp-workflows) for practical first prompts
- [MCP best practices](https://zentrik.ai/docs/integrations/mcp-best-practices) for lighter prompting and realistic limits

## Troubleshooting

### My MCP client still tries an API key

Update the client configuration to use OAuth discovery. Legacy workspace API keys are still valid for `/external/v1`, but they are rejected on `/mcp`.

### I do not see Authenticate yet

Wait a few seconds for the client to finish discovery metadata checks. If it still does not appear, reopen the client and check that the server URL is exactly `https://zentrik.ai/mcp`.

### I received insufficient_scope

The token is valid, but it is missing at least one MCP tool scope needed for the current request. Re-run authorization and request the scopes listed in the `WWW-Authenticate` challenge.

### I authenticated into the wrong workspace

Run the OAuth flow again and choose the correct workspace. Zentrik MCP access tokens are bound to one workspace at a time.

## Related guides

- [Set up Zentrik in ChatGPT](https://zentrik.ai/docs/integrations/mcp-chatgpt)
- [Start with Zentrik MCP](https://zentrik.ai/docs/integrations/mcp-workflows)
- [Use Zentrik MCP well](https://zentrik.ai/docs/integrations/mcp-best-practices)