Zentrik Agent Plugin

Connect an AI agent to the product decisions, customer evidence, and product context your team keeps in Zentrik.

The repository root follows the Agent Plugins v1 format and bundles the production Zentrik Streamable HTTP MCP server with four reusable workflows:

• brief-product-work: ground planning, scoping, building, and review in the workspace's existing evidence.
• set-up-product-workspace: inspect a workspace and apply only an approved, bounded setup delta.
• import-product-evidence: capture supplied calls, tickets, reviews, feedback, or research with an approval step before writes.
• manage-product-study: plan, refine, and review a private Study through evidence-grounded draft findings.

The plugin contains no credentials and no local executable code. The MCP server authenticates through Zentrik OAuth, binds each connection to one selected workspace, and enforces the connected user's role and granted scopes.

Choose the right surface

• Portable Agent Plugins clients load the repository root and connect to the broad https://zentrik.ai/mcp endpoint.
• Codex loads the OpenAI adapter at com.openai/zentrik, which maps to the registered Zentrik app available to ChatGPT and Codex. Both packages include the four workflows above; the app uses the narrower ChatGPT tool profile.
• MCP-only clients can connect directly to https://zentrik.ai/mcp and complete Zentrik OAuth.
• Server-to-server automation should use the versioned REST API at /external/v1 with a scoped API key. The REST API is documented separately and is not bundled in this plugin.

See SURFACES.md for the compatibility contract and the reason the available tools differ by surface.

GitHub Copilot and VS Code

The server is published in the Official MCP Registry as ai.zentrik/mcp. To add it by hand, put this in .vscode/mcp.json (or your user MCP configuration):

{
  "servers": {
    "zentrik": {
      "type": "http",
      "url": "https://zentrik.ai/mcp"
    }
  }
}

Start the server, sign in to Zentrik in the browser window that opens, and pick the workspace this connection may use. Then ask Copilot in agent mode things like:

• "What do customers say about bulk export? Cite the evidence."
• "Brief me on the initiative behind this branch before I change it."
• "Which open ideas have the most customer evidence this quarter?"

Writes, such as importing evidence or updating a record, are limited to your Zentrik role and the scopes you granted.

Codex CLI

This repository includes a Codex marketplace catalog for local, Git, and team distribution:

codex plugin marketplace add Zentrik-AI/agent-plugins
codex plugin add zentrik@zentrik-agent-plugins

The repository root remains the portable artifact. The marketplace entry points Codex at the client adapter under com.openai/zentrik; this keeps OpenAI-specific metadata out of the portable manifest.

OpenAI directory publication

Run python3 scripts/package-openai-plugin.py --output zentrik-openai.zip to package the adapter for the existing directory listing. The packager derives its technical name from the registered app mapping. The ZIP includes the four skills, assets, license files, and the existing ChatGPT MCP endpoint in the portable manifest format. Keep that endpoint in every update; the existing listing owns its OAuth configuration, so the ZIP must not override it. The CLI marketplace keeps the zentrik name and registered app mapping. Upload the ZIP as a new metadata and skills version in the existing listing; server-only changes use continuous MCP review.

Requirements

A Zentrik account and membership in at least one Zentrik workspace. No API key is required.

Documentation and support

• Codex setup: https://zentrik.ai/docs/integrations/mcp-codex
• ChatGPT setup: https://zentrik.ai/docs/integrations/mcp-chatgpt
• MCP setup: https://zentrik.ai/docs/integrations/mcp
• MCP workflows: https://zentrik.ai/docs/integrations/mcp-workflows
• Support: support@zentrik.ai

License

Apache-2.0. See the LICENSE file in the distribution repository.