Connect other agents
The quickstart connects Claude Code, with a note for VS Code agent mode. Those two have been run against Retiform. This page covers three more agents.
These steps are in beta. They are written from each product's own documentation and from how Retiform's sign-in works, and nobody has yet run them against Retiform from start to finish. If you try one, please tell us what happened, whether it worked or not, at the support address at the foot of this page. Your report is how a step here moves out of beta.
What every agent needs
- Your MCP address. Your console's Connect your agent panel shows it; it ends in
/mcp. Enter it exactly as shown. - No key. Retiform uses sign-in, not an API key. The agent opens a browser page, and there you sign in with the code Retiform emails you, approve the agent in Retiform's console, and confirm once more. The quickstart's step 2 describes those three screens.
- Room for one more agent client. Each agent you connect counts as one agent client of your tenant, and the trial allows a fixed number of them (Limits). You can revoke one of yours that you no longer use from the console, and a tenant administrator can revoke anyone's.
If sign-in does not start, or the agent says it cannot register, see "If it does not work" below.
claude.ai
Custom connectors are in every Claude plan. On the Free plan you can add one. On a Team or Enterprise plan, an owner adds the connector for the organisation, and each member then connects with their own sign-in.
- In claude.ai, open Customize, then Connectors. (Anthropic's documentation has also called this Settings, then Connectors.)
- Choose Add, then Add custom connector.
- Enter a name, such as Retiform, and your MCP address. Choose Continue.
- Leave the authentication options as detected. If you are asked how Claude should identify itself, choose Register automatically.
- Choose Add, then Connect, and sign in to Retiform as described above.
- In a chat, choose +, then Connectors, and switch Retiform on for that conversation.
On a Team or Enterprise plan, an owner does steps 2 to 4 under Organization settings, then Connectors. Members then find Retiform under Customize, then Connectors, and choose Connect.
Things to know:
- Claude asks before it uses a tool, unless you set a tool to "Always allow" in the connector's settings. Recording memory and uploading a document change your tenant, so consider leaving those on "Needs approval".
- The connection comes from Anthropic's servers, not from your computer.
- If you remove the connector and add it again, Claude registers a new agent client, which takes another of your tenant's agent client places. Revoke the old one on the console's agents list, which shows when each was last used. If you are at the limit, the approval page offers to revoke a left-over one and approve the new one.
Anthropic's instructions: "Getting started with custom connectors using remote MCP" at support.claude.com, and claude.com/docs/connectors/custom/remote-mcp.
Claude Desktop
Claude Desktop uses the same connectors as claude.ai. They belong to your Claude account, not to the application, so a connector you added on the web is already there.
- In Claude Desktop, open Customize, then Connectors.
- If Retiform is listed, choose Connect if it asks. If it is not, add it as in the claude.ai steps.
Do not add Retiform to claude_desktop_config.json. That file is for programs that run on your own
computer, and Retiform is a remote service.
Cursor
Cursor reads its MCP servers from a file: .cursor/mcp.json in a project, or ~/.cursor/mcp.json for all
your projects.
-
Create or open the file and add Retiform, with your own MCP address:
{ "mcpServers": { "retiform": { "url": "<your MCP address>" } } } -
In Cursor, open Customize in the sidebar and switch the
retiformserver on. -
Cursor opens your browser to sign in. Sign in to Retiform as described above.
-
Back in Cursor, check that Retiform's tools are listed under the server.
Things to know:
- Cursor asks before it uses a tool by default.
- Do not put a key or a token in the file. None is needed, and a project file is often committed.
- Cursor's cloud agents sign in separately from the desktop application, and may count as another agent client.
Cursor's instructions: cursor.com/docs/context/mcp.
If it does not work
| What you see | What to check |
|---|---|
| The agent says the address is wrong, or nothing happens | The address must be the one from your invite, ending in /mcp, with https:// in front |
| No sign-in page opens | Remove the server or connector and add it again. In claude.ai, choose "Register automatically" when asked how Claude identifies itself |
| You signed in, but the agent shows no tools | You may not have approved the agent in Retiform's console. Open the console and look for a waiting approval |
| The console says the tenant has no room for another agent client | The approval page offers to revoke an agent client of yours that is no longer used and approve the new one. If none is yours, ask a tenant administrator to revoke one, then connect again |
A tool fails with INSUFFICIENT_SCOPE |
The agent was approved without that permission. Revoke it in the console and connect again, ticking what it needs |
| A tool or option the changelog describes is missing | Your agent may keep the tool list it read before a release. Quit the agent and open it again: a reconnect alone may not refresh it |
| Something else | Troubleshooting lists each error. Then write to us and say which agent you used and what it showed |
Other agents
Any agent that supports the Model Context Protocol over HTTP with sign-in in the browser should be able to connect in the same way: give it the MCP address, and sign in when it asks. Retiform does not accept an API key in place of sign-in. We have not tried agents other than those named on this page and in the quickstart.