Run Your Envelope AI Agents From Claude and ChatGPT
June 12, 2026 · 5 min read
There's a version of this we've wanted for a while: you're in a Claude conversation, a ticket comes up, and instead of switching tabs to trigger your support triage team, you just say "run the support triage team on this" — and it runs.
That's what the Envelope MCP server makes possible.
What we shipped
Envelope now runs a hosted MCP server at mcp.openenvelope.org. Add it as a connector in Claude.ai or ChatGPT and your Envelope workspace appears as a set of tools in your AI conversations.
The tools cover the operations you'd want to trigger from a conversation:
- List your installed teams — see what's running, what's paused, what's available
- Run a team — trigger any installed team with optional input
- Continue or reject a gate — handle human-in-the-loop review checkpoints inline
- Cancel a run — stop a run in progress
- Query run history — check what ran recently and what it produced
- Browse templates — see your template library
One server. Works with Claude.ai and ChatGPT. No separate integration per client.
How to connect
Claude.ai:
- Go to claude.ai → Customize → Connectors → Add custom connector
- Paste
https://mcp.openenvelope.org/api/mcp - Envelope will redirect you to sign in with your Envelope account
- Once connected, the tools appear in Claude's connector list
ChatGPT:
- Go to chatgpt.com → Settings → Apps → New App
- Paste the server URL:
https://mcp.openenvelope.org/api/mcp - ChatGPT uses OAuth — you'll be redirected to authorise with Envelope
- Approve access and your teams appear as available tools
The authentication uses a standard OAuth 2.0 flow with PKCE. Your Envelope credentials are never shared with the AI client — the client gets a scoped access token that it uses to call the Envelope API on your behalf.
What it looks like in practice
You're talking to Claude about a support backlog. You ask it to run your triage team on a batch of tickets. Claude calls run_team, passes the context, and returns the result — classifications, drafted responses, escalation flags — back in the same conversation.
Or you're working through a code review in ChatGPT. A gate is waiting on your PR summary team. You tell ChatGPT "approve the pending gate on the PR summary team" — it calls continue_gate and the pipeline resumes.
The key shift: your AI teams stop being something you manage in a separate tool and start being available wherever you're already working.
Claude Code too
If you use Claude Code in the terminal, the Envelope MCP skill is available on buildwithclaude.com — search for envelope-team. The skill gives Claude Code full knowledge of the .envelope.json schema so it can generate and validate team definition files from natural language.
The remote MCP server and the Claude Code skill are complementary. The skill helps you design and define teams. The server lets you run and manage them from any conversation.
What's next
The MCP surface will expand as the API does. If there's a tool you'd find useful — something that maps to an existing API operation — let us know at [email protected].
The full MCP reference, including all tool signatures and parameters, is at openenvelope.org/docs/mcp.
Frequently Asked Questions
What is the Envelope MCP server?
The Envelope MCP server is a hosted server at mcp.openenvelope.org that exposes your Envelope workspace as tools inside AI conversations. Connect it once and you can run agent teams, handle approval gates, and check run history without leaving Claude or ChatGPT.
How do I connect it to Claude?
In Claude.ai, go to Settings → Integrations and add mcp.openenvelope.org/api/mcp as a connector. Your Envelope teams will appear as available tools in any Claude conversation immediately after.
Which AI clients does it support?
Claude.ai and ChatGPT. Both support the MCP protocol. One server URL works for both — no separate integration per client.
What can I do through the MCP integration?
List your installed teams, run a team with optional input, continue or reject human-in-the-loop review gates, cancel a run in progress, query recent run history, and browse your template library — all from within the conversation.