> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.cessy.ai/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.cessy.ai/_mcp/server.

# MCP Reference

> Compare Cessy Design MCP and Runtime MCP surfaces for builder and runtime agent workflows.

Cessy exposes different MCP servers for different jobs. Choose the narrowest server that matches the task.

## Design MCP

Design MCP is for builder work:

* inspect apps and DomainConfigs
* read reference knowledge
* create workspaces
* edit model artifacts
* validate changesets
* preview pages
* inspect errors
* promote approved changes

Endpoints:

```text
/api/design-mcp-readonly?tenantId=<tenant-id>&appId=<app-id>
/api/design-mcp?tenantId=<tenant-id>&appId=<app-id>
/api/design-mcp
```

Use read-only for onboarding, review, and discovery. Use full Design MCP only for trusted model-changing sessions.

## Runtime MCP

Runtime MCP is for published app operation:

* command tools
* projection tools
* event import and projection rebuild tools
* channel tools
* user and DM tools
* page refresh tools

Tool names and schemas come from the published DomainConfig.

Endpoint:

```text
/api/mcp/<tenant-id>/<app-id>
```

Runtime MCP requires bearer authentication with the `runtime` scope, or an app-scoped agent service token. For workspace-selected runtime calls, use `workspaceId=<workspace-id>` or `X-CES-Workspace-ID`; the older `env` query parameter is not supported.

## Client support

Codex supports Streamable HTTP MCP servers through `codex mcp` and `~/.codex/config.toml`. Claude and ChatGPT can connect through custom connector flows by using the app-scoped Remote MCP URL.

See [Connect Design MCP](/start/connect-design-mcp).

## Tool design rule

MCP tools should expose intentional workflows, not unmodeled data access. A tool should make authority, target app, target environment, and side effects clear.