MCP reference
Every MCP tool exposed by the IMMA AI MCP server, how it maps to the REST API, and setup guides for Claude, Cursor, ChatGPT and n8n.
Preview
Preview: the API and MCP server are in private beta; details may change.
The IMMA AI MCP server lives at https://api.getimma.com/mcp (Streamable HTTP). It is the primary way an AI agent, such as Claude, ChatGPT or Cursor, posts and monitors social media on a customer's behalf: the customer's own AI subscription writes captions, drafts a content calendar, or generates images, and MCP is the "hands" that upload, schedule and route the result through approval. IMMA AI does not generate images or video itself; it only stores and publishes what the agent's own AI produced.
If you have not connected a client yet, start with the MCP quickstart for the shortest path to a first draft post. This section goes deeper: a setup page per client, the full tool reference, and troubleshooting.
MCP is a thin wrapper over the REST API
Every MCP tool calls the same core logic as a REST endpoint, so the rules are identical no matter which one you use:
| MCP tool | REST equivalent |
|---|---|
create_post | POST /v1/posts |
validate_post | POST /v1/posts/validate |
create_posts_batch | POST /v1/posts/batch |
get_calendar | GET /v1/calendar |
upload_media | POST /v1/media |
get_post / list_posts | GET /v1/posts/{id} / GET /v1/posts |
update_post / cancel_post | PATCH /v1/posts/{id} / POST /v1/posts/{id}/cancel |
get_analytics | GET /v1/analytics/posts/{id} or /v1/analytics/accounts/{id} |
list_comments / reply_comment | GET /v1/comments / POST /v1/comments/{id}/reply |
See the API reference for the underlying request and response shapes; the tool reference below documents the same fields from the MCP side.
Which client uses which auth
| Client | Auth method | Notes |
|---|---|---|
| ChatGPT (Developer mode app) | OAuth 2.1 login | Needs Developer mode (Pro, Plus, Business, Enterprise or Education). Custom connectors cannot accept a static key. |
| Claude.ai / Claude Desktop (Customize > Connectors) | OAuth 2.1 login | Works on Free, Pro, Max, Team and Enterprise (Free: one custom connector). Click Connect, log in, no key copying. |
| Claude Desktop (mcp-remote bridge) | Bearer API key | Advanced/alternative fallback for machines that cannot complete an interactive login. |
| Claude Code | Bearer API key | claude mcp add with an Authorization header. |
| Cursor | Bearer API key | mcp.json with an Authorization header. |
| n8n / scripts | Bearer API key | Header Auth node, or any HTTP client. |
Safety rules, enforced the same way as the REST API
- TikTok always goes through approval or inbox. An agent can never choose a TikTok privacy level on a user's behalf.
create_postandcreate_posts_batchdefault every TikTok target to a hosted approval link (approval: { mode: "link" }), or you can sendtiktok.mode: "inbox"to deliver a draft to the creator's TikTok inbox instead. There is no MCP input that skips this. See Platform rules. - No platform defaults. Interaction toggles (comment, duet, stitch) always start off; nothing is preselected for the account owner.
- Rules live in
packages/core, not in any one client. MCP, the REST API and the dashboard composer enforce the same validation, so there is no way to bypass consent by switching to a different integration path. - Scoped access. Scopes (
posts:write,posts:read,accounts:read,accounts:write,analytics:read, and so on, the same scopes as the REST API, see Authentication) control which tools even appear intools/list, whether you connected with an OAuth login or a Bearer API key. A read-only grant never sees write tools. - Content is never treated as instructions. Text inside a caption or a comment is data, not a command; the MCP server only executes explicit tool calls from your client.
Setup guides
Reference
- Tools: every tool, its arguments, an example call and result, and the rules it enforces.
- Troubleshooting: common errors and what to do about them.