# AI agent and MCP client compatibility

Source: [AI agent and MCP client compatibility](https://typeroll.com/docs/getting-started/client-compatibility/)

> Connection options and Typeroll verification status for Claude Desktop, Claude Code, Cursor, VS Code, ChatGPT, Cline and Zed.

Typeroll CMS connects to AI agents through MCP and the REST API. It does not require a particular AI model or client. An MCP client must support the transport, authentication and tool calls used by the connection you choose.

## Connection options

[Section titled “Connection options”](https://typeroll.com/docs/getting-started/client-compatibility/#connection-options)

* **Remote:** MCP Streamable HTTP at `https://app.typeroll.com/api/mcp`, or `https://<your-portal-host>/api/mcp` for self-hosting. Use OAuth consent or an explicit bearer header where the client supports it.
* **Local:** launch `npx -y @typeroll/mcp-server` over stdio with the environment variables in [Connect your AI agent](https://typeroll.com/docs/getting-started/mcp-server/#local-mcp-process-stdio). A stdio process targets one Site; set `TYPEROLL_SITE_ID` when the key can access multiple sites.
* **Direct API:** an agent can call the authenticated REST API without MCP. Reading the [public documentation](https://typeroll.com/docs/llms.txt) requires neither a key nor MCP.

A client that only supports older HTTP+SSE connections is not automatically compatible with the Streamable HTTP endpoint. Configuration schemas, available features and workspace permissions also differ between clients.

## Client status

[Section titled “Client status”](https://typeroll.com/docs/getting-started/client-compatibility/#client-status)

Provider documentation reviewed on **September 11, 2026**. The connection options below are based on those providers’ documentation. **A version-specific Typeroll end-to-end test has not yet been recorded for these named clients.** They are connection candidates, not a list of certified integrations.

| Client                                                                                           | Connection to configure                            | Typeroll client test |
| ------------------------------------------------------------------------------------------------ | -------------------------------------------------- | -------------------- |
| [Claude Desktop](https://typeroll.com/docs/getting-started/client-compatibility/#claude-desktop) | Remote custom connector                            | Not yet recorded     |
| [Claude Code](https://typeroll.com/docs/getting-started/client-compatibility/#claude-code)       | Remote HTTP or local stdio                         | Not yet recorded     |
| [Cursor](https://typeroll.com/docs/getting-started/client-compatibility/#cursor)                 | Remote HTTP or local stdio                         | Not yet recorded     |
| [VS Code](https://typeroll.com/docs/getting-started/client-compatibility/#vs-code)               | Remote HTTP or local stdio in its agent tools      | Not yet recorded     |
| [ChatGPT](https://typeroll.com/docs/getting-started/client-compatibility/#chatgpt)               | Remote custom MCP app, subject to workspace access | Not yet recorded     |
| [Cline](https://typeroll.com/docs/getting-started/client-compatibility/#cline)                   | Remote Streamable HTTP or local stdio              | Not yet recorded     |
| [Zed](https://typeroll.com/docs/getting-started/client-compatibility/#zed)                       | Remote or local context server                     | Not yet recorded     |

Typeroll’s automated checks exercise its server tools and OAuth implementation. Those checks do not substitute for completing authentication and tool calls in a particular third-party application.

### Claude Desktop

[Section titled “Claude Desktop”](https://typeroll.com/docs/getting-started/client-compatibility/#claude-desktop)

Use a remote custom connector with the hosted endpoint. The remote connection is made from the provider’s infrastructure, so the portal must be reachable there; your computer’s `localhost` is not a public server address. Follow [Claude’s custom connector instructions](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp).

### Claude Code

[Section titled “Claude Code”](https://typeroll.com/docs/getting-started/client-compatibility/#claude-code)

Claude Code documents both HTTP and stdio MCP connections. Use its MCP setup command for the selected transport and its authentication flow for a remote server. Follow [Claude Code’s MCP guide](https://code.claude.com/docs/en/mcp).

Typeroll’s optional `init` command currently creates `.mcp.json` and `.claude/skills/` files. That scaffold is client-specific; other agents can use the MCP connection and `read_guide`, `list_skills` and `read_skill` without it.

### Cursor

[Section titled “Cursor”](https://typeroll.com/docs/getting-started/client-compatibility/#cursor)

Use Cursor’s MCP settings and its own `mcp.json` schema. Its documentation covers stdio commands, remote connections and OAuth. Follow [Cursor’s MCP guide](https://prod.cursor.com/docs/mcp); do not assume a different editor’s configuration file can be copied unchanged.

### VS Code

[Section titled “VS Code”](https://typeroll.com/docs/getting-started/client-compatibility/#vs-code)

Use VS Code’s MCP server management for agent tools. Its configuration uses a `servers` map, which differs from clients using `mcpServers`. Follow [VS Code’s MCP setup guide](https://code.visualstudio.com/docs/agent-customization/mcp-servers) for adding a server and approving access.

### ChatGPT

[Section titled “ChatGPT”](https://typeroll.com/docs/getting-started/client-compatibility/#chatgpt)

Use a custom remote MCP app where your account and workspace permit it. Check read and write tool availability separately: access to a connector does not establish that editing and deployment tools are enabled. Developer-mode and administrator requirements vary by plan. Follow [ChatGPT’s MCP app instructions](https://help.openai.com/en/articles/12584461-developer-mode-and-full-mcp-connectors-in-chatgpt-beta). The stdio npm command is not a remote app URL.

### Cline

[Section titled “Cline”](https://typeroll.com/docs/getting-started/client-compatibility/#cline)

Cline documents both stdio and remote connections. For the hosted Typeroll endpoint, explicitly select **Streamable HTTP**; do not leave it on the legacy SSE transport. Its remote configuration supports bearer headers. Follow [Cline’s MCP guide](https://docs.cline.bot/mcp/mcp-overview).

### Zed

[Section titled “Zed”](https://typeroll.com/docs/getting-started/client-compatibility/#zed)

Zed calls these connections context servers and configures them under `context_servers`. It documents local commands, remote URLs, bearer headers and OAuth when no Authorization header is configured. Follow [Zed’s MCP guide](https://zed.dev/docs/ai/mcp).

## Verify your connection

[Section titled “Verify your connection”](https://typeroll.com/docs/getting-started/client-compatibility/#verify-your-connection)

Start with a read-only request: identify the accessible Site and Version, read the site capabilities and read a page. For remote multi-site connections, use `list_sites` to choose the Site; a stdio connection is bound at startup.

If you want to verify editing, use a disposable test Site. Create a draft, read it back, make a small change and inspect its preview. Test saving, deletion and publishing only when you intend those actions. A successful read alone does not prove that writes, authentication renewal or deployment work in that client.

When recording a verified client, include the client version, OS, Typeroll version and environment, transport, authentication method, test date and actual operations exercised. Record known restrictions alongside the result. Never include keys, tokens or private customer content in a compatibility report.