Customer evidence in the AI tools your team already uses
Connect Claude, ChatGPT, Codex, Cursor, or any MCP client to one Zentrik server. Each tool reaches the same signals, opportunities, and decisions, inside the permissions your team already set.
Listed in the Claude Connectors Directory and the Plugins Directory for ChatGPT and Codex. Published to the Official MCP Registry.
Remote MCP server
zentrik.ai/mcp
Start where your team already works
PMs ask in Claude or ChatGPT, engineers build in Claude Code, Codex, or Cursor. Every route signs in the same way and reads the same workspace.
Claude and Cowork
Claude Connectors Directory
Connect Zentrik from the directory, approve one workspace, and ask Claude about evidence, opportunities, and Studies. Writes can wait for your approval under the tool permissions you set in Claude.
ChatGPT
Plugins Directory
Install Zentrik from the Plugins Directory to search customer evidence, inspect the source trail behind an opportunity, and save new customer notes from a conversation.
Claude Code
Zentrik plugin marketplace
Install the plugin to bring the connection and four product workflows into Claude Code, so the agent reads the decision and its evidence before it writes code.
claude plugin marketplace add Zentrik-AI/claude-plugins claude plugin install zentrik@zentrik
Codex
Plugins Directory and Agent Plugin
Install Zentrik from the Plugins Directory in the ChatGPT desktop app, or add the Zentrik Agent Plugin to Codex CLI with the same workflows.
codex plugin marketplace add Zentrik-AI/agent-plugins codex plugin add zentrik@zentrik-agent-plugins
Cursor
MCP settings
Add Zentrik to mcp.json in your project or global settings, choose Connect, and approve a workspace in the browser.
{
"mcpServers": {
"zentrik": { "url": "https://zentrik.ai/mcp" }
}
}Any MCP client
Streamable HTTP with OAuth
Point any client that supports remote MCP servers at https://zentrik.ai/mcp. Zentrik handles sign-in, workspace choice, and scopes. The server is published to the Official MCP Registry as ai.zentrik/mcp.
https://zentrik.ai/mcp
Every record behind a product decision
More than 100 tools cover the product graph, so an agent can follow a request from the customer signal to the initiative it shaped.
- evidenceSignals and their sources
- Sales calls, support tickets, research, reviews, and imports, each kept with the source it came from.
- patternsInsights and opportunities
- The problems that repeat across customers and the opportunities they make legible, with links back to every signal.
- proposalsIdeas and Studies
- Proposed moves, their planning questions, and private Studies with response coverage and draft findings.
- deliveryInitiatives and documents
- Approved bets with their briefs, PRDs, user stories, and the evidence and constraints that travel into delivery.
- contextProducts, personas, accounts, and OKRs
- The product structure, customers, and goals an agent needs to judge whether a request fits.
Named workflows for the jobs that repeat
Claude Code and Codex install all four as skills. Full MCP clients also list workspace setup and evidence import as ready-made prompts.
- 01/zentrik:brief-product-work
Brief product work
Briefs the work in front of you with the decisions, customer evidence, and open questions the workspace already holds, each attributed to a record ID.
- 02/zentrik:set-up-product-workspace
Set up a product workspace
Inventories the workspace, asks for the smallest missing set of sources, and applies only the changes you approve.
- 03/zentrik:import-product-evidence
Import product evidence
Captures calls, tickets, reviews, and research as evidence, one coherent source at a time, with a preview before batches.
- 04/zentrik:manage-product-study
Manage a product Study
Plans a private Study from an idea or initiative, reviews response coverage, and drafts evidence-grounded findings for your team to review.
How access works
- OAuth 2.0 with PKCESign in with Zentrik instead of sharing a key
- Clients authenticate through OAuth with PKCE and dynamic client registration, so nobody pastes a long-lived API key into an AI tool.
- one workspaceEvery connection is bound to one workspace
- The person connecting chooses the workspace during consent. Reconnecting is the only way to switch.
- role and scopeYour Zentrik role still applies
- Viewers get read access. Editors, Admins, and Owners can create and update records within the scopes they granted.
- annotated toolsClients know which tools change data
- Every tool carries a title and a read-only or destructive annotation, which lets Claude, ChatGPT, and other clients ask before a write runs.
Questions about the Zentrik MCP server
What is the Zentrik MCP server?
It is Zentrik's remote Model Context Protocol server at https://zentrik.ai/mcp. AI tools connect to it with OAuth to search and update the customer evidence, opportunities, ideas, Studies, and initiatives in a Zentrik workspace.
Which AI tools work with Zentrik?
Claude, Cowork, and Claude Code; ChatGPT and Codex; Cursor; and any client that supports remote MCP servers over Streamable HTTP with OAuth.
Where is Zentrik listed?
Zentrik is listed in the Claude Connectors Directory and in the Plugins Directory shared by ChatGPT and Codex, and published to the Official MCP Registry as ai.zentrik/mcp. The Claude Code plugin and the Codex Agent Plugin install from Zentrik's public repositories.
Why does ChatGPT offer a smaller set of tools than Claude?
ChatGPT and Codex use the tool set published with the current Plugins Directory version, which focuses on searching evidence, inspecting opportunities, and capturing notes. Claude and other MCP clients connect to the full server, and new tools reach ChatGPT with each new directory version.
Do I need an API key?
No. MCP clients sign in with OAuth. For server-to-server automation, the Zentrik REST API accepts a scoped API key instead.
Can an agent decide what we build?
No. Zentrik returns evidence and prepares work. Writes need a capable role, clients can ask for approval before they run, and people still own the product decision.
Give every agent the reason behind the request
Create a workspace, connect the tool your team opens first, and check the source records before you decide.