Connect your AI client to ConvOps over MCP
ConvOps, the operations layer for AI agents, is a remote MCP server at https://mcp.convops.app/. Add that URL to Claude Code, Claude Cowork, OpenAI Codex, ChatGPT, Cursor or OpenCode, sign in once in the browser, and your tasks, workflows, approvals and team memory arrive as tools in that client. ConvOps runs no AI models.
https://mcp.convops.app/examplefrom Claude Code
Which AI clients can connect to ConvOps?
These six have a step-by-step guide with the exact command or config, the sign-in, and a governed task running in that client. Any other client that can add a remote MCP server over HTTP connects with the same URL.
- AnthropicClaude CodeOne command in the terminal. Sign in with /mcp. setup guide
- AnthropicClaude CoworkCowork, Claude desktop and claude.ai. One custom connector. setup guide
- OpenAICodexCLI and IDE extension. One command, then codex mcp login. setup guide
- OpenAIChatGPTDeveloper mode on the web. Create an app with the URL. setup guide
- AnysphereCursorOne entry in mcp.json, global or per project. setup guide
- open sourceOpenCodeA remote entry in opencode.json, then opencode mcp auth. setup guide
How does connecting an AI client to ConvOps work?
Three moves, the same in every client. Only the first one looks different from client to client.
- 1
Add the server
Give your client the server URL: a command in Claude Code or Codex, an entry in mcp.json or opencode.json, or a connector form in Claude Cowork and ChatGPT. No agent to deploy, no data to import.
- 2
Sign in once
On first use the client opens a browser to sign in with ConvOps. OAuth identifies you and your workspace, so there are no API keys to copy.
- 3
Say the process
Describe how the work should go, in plain language. It compiles into a workflow with steps, gates and decisions, and your AI runs it from this same client.
After that, the steps, gates and decisions come from the workflow engine, and approvals and the audit trail from governance. New to the terms? Start with what MCP is, or see why a workflow beats a long instruction file in how to make Claude Code follow a process.
Already using a workflow tool? The ConvOps comparisons show where it fits next to Claude Code Dynamic Workflows, Archon, Claude Task Master, LangGraph, Temporal and HumanLayer, and the changelog lists what shipped in the server you are connecting to.
How does setup differ between clients?
The URL is the same everywhere; where it goes and how you sign in differ. Checked against each vendor's docs in October 2026, and each guide links them, because menu names change between versions.
| Client | Add it with | Config lives in | Sign-in | API key path |
|---|---|---|---|---|
| Claude Code | claude mcp add command | ~/.claude.json or .mcp.json | OAuth via /mcp | --header flag |
| Claude Cowork | Custom connector in settings | Customize, Connectors | OAuth when you connect | Not used (OAuth) |
| Codex | codex mcp add command | ~/.codex/config.toml | codex mcp login | --bearer-token-env-var |
| ChatGPT | App in developer mode | ChatGPT settings | OAuth when you create the app | Not used (OAuth) |
| Cursor | mcp.json entry | ~/.cursor/mcp.json or .cursor/mcp.json | OAuth on first connect | headers in mcp.json |
| OpenCode | opencode.json entry | opencode.json (project or global) | opencode mcp auth | headers with oauth: false |
Can you connect with an API key instead of signing in?
Yes, for automation or a client without OAuth. Create a key in the ConvOps app and send it as an Authorization Bearer header. Keys start with ntk_, belong to your user, can be narrowed to specific workspaces, are shown once and stored hashed. Each guide shows the header setting for its client.
Authorization: Bearer ntk_...What else do people ask about connecting AI clients?
Which AI clients work with ConvOps?
Claude Code, Claude (web, desktop and Cowork), OpenAI Codex, ChatGPT in developer mode, Cursor and OpenCode each have a setup guide here, checked against each vendor's docs in October 2026. Any other client that can add a remote MCP server over HTTP can connect with the same URL.
Do I need an API key to connect?
No. Every client on this page signs in with OAuth in the browser on first use. API keys exist for automation and for clients without OAuth: create one in the ConvOps app and send it as an Authorization Bearer header.
Does ConvOps run the AI model?
No. Your client runs the model and does the work. ConvOps holds the process: it hands over one step at a time, checks the gate when the agent asks to advance, and records what happened.
Can my team use different clients on the same workflow?
Yes. The task, its workflow and its history live in your ConvOps workspace, not in the client, so one person can start a task in Claude Code and another can pick it up in Cursor at the step where it stopped.
What does it cost to connect?
Connecting is free, and every plan, Free included, allows unlimited AI clients. Paid plans are Pro at $79 a month ($66 billed annually) and Business at $299 a month ($249 annually); Enterprise starts from $2,500 a month. Your AI client is billed separately by its vendor.
What if I belong to several workspaces?
With one workspace, it is selected when you sign in. With several, the AI passes a workspace on each call, so one connection can work across all of them.