# Operate Iterada from your agent

> Connect Claude Code, Codex, Cursor or any MCP agent to Iterada so it starts sessions, follows them, answers their questions and triages issues for you.

Operate Iterada from the agent you already work in. Connected, it hands your projects work, follows each session, answers the questions a session asks, accepts a spec and triages what the quiet-hours passes found, all in the conversation you are already having with it. Iterada's own agents still write the code, and every change still reaches your product the way your project delivers it.

Iterada is a remote MCP server, the standard every major coding agent reads tools from, so one connection works the same in each of them.

## What a connected agent can do

| What it asks for | What happens |
| - | - |
| List projects and sessions | It sees each session's state - working, waiting for an answer, finished, failed - and where its change is: on its own branch, merged or live. |
| Start a session | Iterada's coding agent makes the change, checks it and delivers it under the project's rules. Ask for a spec first on anything larger than one obvious change. |
| Follow a session | It reads the latest reply, the question and its options when a session is waiting, and a proposed spec. It can wait for a turn to finish instead of asking again and again. |
| Answer, steer or stop | It replies in the session, accepts a proposed spec, stops a running turn, or continues one it stopped. |
| Triage issues | It reads what the quiet-hours passes found, starts a fix on one, or dismisses it. |
| Read artifacts | It reads the reports, documents and datasets your project's agents published. |

Three things stay yours: production deploys, which you start from **Production**; a merge your project leaves to you; and secrets, settings, billing and your account. No connection can touch them.

## Connect an agent

1. Open your **Account** page from the avatar in the console.
2. Under **Agent access**, name the agent, for example Claude Code, and choose **Operate** or **Read only**. Leave **All projects**, or pick one project to keep the agent to it.
3. Press **Create token**, then **Copy** the token and press **Done**. It is shown only once.
4. Add Iterada to your agent as a remote MCP server: the address shown under **Agent access**, with the header Authorization: Bearer followed by your token.

In Claude Code, that is one command in a terminal, with your token in place of YOUR_TOKEN:

:::prompt In a terminal
claude mcp add --transport http iterada https://iterada.com/mcp --header "Authorization: Bearer YOUR_TOKEN"
:::

| Agent | Where to add it |
| - | - |
| Claude Code | The command above, or the same server in a project's MCP settings. |
| Codex | Its MCP settings, as a streamable HTTP server whose bearer token is read from an environment variable holding your token. |
| Cursor, Windsurf, VS Code | MCP settings, as a remote server with the address and an Authorization header. |
| Gemini CLI, OpenCode | MCP settings, as a remote HTTP server with the same header. |
| Claude Desktop | Through the mcp-remote bridge, given the address and the same header. |
| Your own script | Any MCP client library, or plain JSON-RPC posted to the address with the header. |

:::note
Claude.ai and ChatGPT add connectors by signing in to them rather than with a token, which Iterada does not offer yet.
:::

## Ask for work

Then talk to your agent as you would to a colleague who has Iterada open:

:::prompt Example request to your agent
In Iterada, start a session in my shop project that adds a coupon field at checkout. Ask for a spec first, show me the spec, and once I agree, accept it and tell me when the change is live.
:::

When a session asks a question that is yours to decide, the agent brings it to you rather than guessing, unless you already told it the answer. Every session it starts is in the console like any other, so you can open it, read along or take over at any time. What a connected agent does is recorded in your name and marked "through an agent" in the session's history under **Session details**.

## Remove a connection

Press **Remove** beside the connection under **Agent access**. Its agent is cut off at once; nothing it started is undone.

:::tip
Make one connection per agent. Each shows when it was last used, and removing one leaves the others working.
:::

:::cards
- [Ask for a change](https://iterada.com/docs/ask-for-a-change): what makes a request easy to get right.
- [Spec mode](https://iterada.com/docs/spec-mode): agree on what will be built before it is built.
- [Artifacts](https://iterada.com/docs/issues): what the quiet-hours passes and your agents leave for you.
:::