Skip to content
Start free trial

Which AI and automation tools can prepare my social posts?

Connect Claude, ChatGPT, Cursor or any MCP client, or call the REST API from a script. Each connection gets only the permissions you choose, agent posts wait for your approval by default, and the activity log shows what each one did. The n8n node is not published yet.

Updated

Best for

  • People who want an assistant to prepare posts before a human signs off
  • Developers who want hosted MCP and HTTP access to the same workspace
  • Self-hosted n8n users who already have the node source for a manual install

Not for

  • People expecting an n8n package or CLI already published to npm.
  • Agents that need billing access, account connections or self-approval.
  • Teams needing every discovery request and permission refusal logged.

Pick your integration

  • Claude: Remote MCP connection with OAuth and workspace consent.
  • ChatGPT: Remote MCP connection with OAuth and workspace consent.
  • Cursor: Remote MCP connection with OAuth and workspace consent.
  • n8n: Source access required for manual installation; public REST API as a fallback.

What you get

  • Keep control of publishing: agent posts wait for owner approval by default. A key with the explicit skip-approval permission can bypass that hold; replies are outside the post-approval flow. Source
  • Give each key only the permissions it needs. Keys are shown once, stored hashed and revocable. Agents cannot connect accounts, change billing or approve their own posts. Source
  • The activity log records routed API operations and executed MCP tool calls, with their status. Some permission refusals and MCP discovery requests are outside that record; it is not a complete audit of everything an assistant does. Source
  • Use MCP with clients described in the connection guide, REST for HTTP workflows, a local CLI for shell scripts, and a source-only n8n node for self-hosted automation. A Make HTTP-module guide is also in the agent kit; a dedicated Zapier integration is not built. Source

How it works

  1. Pick the interface you already use

    Choose a supported remote MCP client for a chat workflow, the REST API for HTTP calls, or the local CLI for scripts if you have source access. The unpublished n8n node also requires source access for manual installation; the client guides below explain the connection routes.

  2. Choose one workspace and limited permissions

    An MCP connection asks you to pick the workspace and consent to scopes. API keys are created in Settings > AI agents and are shown once. Give the integration read and draft access before adding scheduling or publishing permissions.

  3. Review the first scheduled post

    Keep default approval on and skip approval off when you want a human handoff. Agent-written posts wait for the workspace owner. Use recorded operation statuses in the activity log to inspect calls that reached the routed API or executed MCP tools.

What it costs

There is no Free plan. Every paid plan includes MCP and API keys. The unpublished CLI requires source access for local installation. Solo is $19 a month for 5 accounts and one seat; Team is $29 for 10 accounts and unlimited seats. Every workspace holds one account per network; several brands on the same network need separate workspaces on Agency. A 7-day trial requires a card when offered; check Pricing for today's availability before signing up.

See plans and today's trial offer

Questions people ask

Can I connect Claude, ChatGPT and Cursor?

Yes. The repository MCP guide describes connection steps for Claude, ChatGPT and Cursor using the hosted server URL. Account and workspace policies still apply, including possible admin restrictions on custom ChatGPT MCP servers.

Do I need to give an agent my social passwords?

No. Agents work with scoped API keys or OAuth consent. They cannot connect social accounts themselves. Connect accounts through the app, then grant the integration access to the workspace you choose.

Can I install the n8n node from npm today?

No. It is not published to npm yet. Manual installation needs source access to build and install the tarball, then API-key credentials. The trigger also needs webhook permission. Without the source, use the public REST API guide instead.

Do replies wait for the same approval as posts?

No. Comment replies go out without the post-approval hold. Only grant reply permission when you want that behavior. Leave it unticked for an integration whose job is preparing drafts.

Can I revoke access later?

Yes. API keys are revocable and stop working when revoked. The MCP guide also explains connected-app consent. Review both permissions and skip approval when you connect an integration, rather than treating scheduling permission as a request to bypass review.

Does the activity log contain every MCP request?

No. Executed MCP tool calls and routed API operations are recorded. Discovery requests and some early permission refusals are outside that record, so it is not a full audit trail of the assistant or its conversation.

Related guides

Sources and limits

  1. Approval and scoped keys agent-kit/README.md:17
  2. Manual n8n install agent-kit/n8n/README.md:11
  3. Recorded API operations src/index.js:273
  4. Recorded MCP tool calls src/features/mcp.js:645
  5. MCP connection guide public/docs-mcp.html:312
  6. Owner-only agent approval src/index.js:1492
  7. Plans and limits src/plans.js:11
  8. Accounts per network src/features/team.js:393
  9. Agency workspaces src/index.js:743

Choose a plan Read the API docs