Antwork MCP Server

The Antwork MCP server lets your AI assistant draft, schedule, publish, and analyze social media posts on your behalf. It speaks the open Model Context Protocol standard, so any MCP-aware client (claude.ai, Claude Code, Cursor, ChatGPT, Windsurf, custom agents) can connect with the same single URL.

What you can do once connected

• ›Draft posts on-brand using your stored voice profile per platform.
• ›Schedule or publish to LinkedIn, X, Threads, Facebook, Instagram, YouTube, TikTok, and Pinterest.
• ›Read engagement metrics, post history, and per-day timeseries for any post.
• ›Manage drafts, scheduled posts, and the media library.
• ›Connect / disconnect social accounts and inspect their health.

Server URL

https://api.antwork.io/mcp

One URL serves every user — there are no per-tenant subdomains. Authentication happens via OAuth on first tool call (see Authentication).

Every supported MCP client follows the same shape: register the server URL, then approve OAuth in your browser when prompted. The exact install affordance differs per client — pick yours below.

Cursor

1. 1.One click — Cursor will open and ask you to confirm the install.

Add to Cursor

One click — your client will open and ask you to confirm the install.

VS Code

1. 1.One click — VS Code will open and prompt you to confirm.

Add to VS Code

One click — your client will open and ask you to confirm the install.

Prerequisite

GitHub Copilot required — MCP servers run in Copilot agent mode. The free Copilot tier works. Learn more ↗

Claude Code

1. 1.Run this command in your terminal.
2. 2.Then in Claude Code, type /mcp to authorize.

Adds Antwork as an HTTP MCP server

$ claude mcp add antwork --transport http https://api.antwork.io/mcp

Prerequisite

Claude Pro, Max, Team, or API key required — Claude Code authenticates against an Anthropic subscription or API key. Learn more ↗

Claude Desktop

1. 1.Open Claude Desktop → Customize → Connectors → Add custom connector.
2. 2.Paste the URL below and click Authorize.

MCP server URL

https://api.antwork.io/mcp

Prerequisite

Claude Pro, Max, Team, or Enterprise required — Custom connectors are not available on the free Claude plan. Learn more ↗

ChatGPT

1. 1.Enable Developer Mode in ChatGPT → Settings → Apps & Connectors.
2. 2.Add a new app: name "antwork", URL below, auth "OAuth".

MCP server URL

https://api.antwork.io/mcp

Prerequisite

ChatGPT Plus, Pro, Business, Enterprise, or Edu + Developer Mode — Custom connectors are gated behind paid plans and require turning on Developer Mode in beta features. Learn more ↗

Windsurf

1. 1.Open Windsurf → Settings → MCP servers → Edit raw config.
2. 2.Paste this snippet inside the mcpServers object.

mcpServers config · Settings → MCP

{
  "mcpServers": {
    "antwork": {
      "url": "https://api.antwork.io/mcp"
    }
  }
}

Gemini CLI

1. 1.Run this command in your terminal.
2. 2.Then run /mcp auth antwork inside the Gemini CLI.

Adds Antwork as an HTTP MCP server

$ gemini mcp add --transport http antwork https://api.antwork.io/mcp

OpenClaw

1. 1.Run this command in your terminal.
2. 2.Then run mcporter auth antwork to authorize.

Adds Antwork via mcporter

$ mcporter config add antwork --url https://api.antwork.io/mcp --auth oauth

The Antwork MCP server implements OAuth 2.1 with PKCE and dynamic client registration (RFC 7591), so MCP clients can connect without you pre-registering each one. Tokens are scoped, revocable, and issued per-client.

OAuth scopes

Each tool requires one of four scopes. The consent screen lists the scopes a client is requesting so you can approve or deny them granularly.

Token lifecycle

• ›Access tokens are short-lived JWTs (5 minute TTL).
• ›Refresh tokens are long-lived and stored server-side. Standard refresh_token grant.
• ›Revoke via the standard /oauth/revoke endpoint (RFC 7009) or from Settings → Connected AIs in your Antwork dashboard. Revocation invalidates the refresh token immediately.
• ›Each MCP client (Claude desktop, Claude Code, Cursor, …) gets its own token. Revoking one client does not affect others.

Discovery endpoints

• ›GET /.well-known/oauth-protected-resource — RFC 9728 resource metadata.
• ›GET /.well-known/oauth-authorization-server — RFC 8414 authorization server metadata.
• ›POST /oauth/register — RFC 7591 dynamic client registration.
• ›POST /oauth/token — authorization-code and refresh-token grants. PKCE S256 required.
• ›POST /oauth/revoke — RFC 7009 token revocation.

Antwork exposes 35 MCP tools across seven categories. Read-only tools never mutate state; destructive tools (publish, delete, disconnect) trigger explicit confirmation in clients that honor destructiveHint annotations.

Identity & workspaces

Social accounts & voice profiles

Posts — read

Posts — write

Posts — publishing

Analytics & planning

Media

Several tools render interactive iframes when invoked from a host that supports the MCP Apps spec (claude.ai, Claude Code ≥ 0.5). The iframe calls back to the same MCP server over the host bridge — no extra auth needed.

• ›list_posts, search_posts → posts-table: filterable list with inline detail modal that edits + publishes without leaving chat.
• ›get_post, create_post, update_post, duplicate_post → post-preview: full post card with per-platform tabs, inline editor, schedule/publish actions, an Add media picker (uploads + library reuse), and per-account "View on X" links once published.
• ›get_calendar → calendar: month / week views of scheduled + published posts.
• ›start_media_upload → upload-zone: workspace-level drop-zone for adding files to the media library without picking a post; supports drag-and-drop, paste, multi-file batches.
• ›list_social_accounts, get_connection_urls, disconnect_social_account → connections-panel: per-account health, reconnect, disconnect; workspace switching via the header propagates to every other iframe.
• ›list_media, get_media → media-gallery: thumbnail grid of the workspace library — reuse media across posts without re-uploading.

Clients that don't support MCP Apps still get the raw tool output — the UI is purely a presentation layer.

"Bearer token required" / 401 on every call

The MCP client hasn't completed OAuth or its refresh token has been revoked. In claude.ai: Settings → Connectors → Antwork → Disconnect, then reconnect. In Claude Code: /mcp → pick antwork → Re-authenticate.

Publish failed on one platform but succeeded on others

publish_post returns per-platform status. The most common cause is a social account whose token expired or was revoked from the platform side. Run list_social_accounts to find unhealthy accounts and reconnect them via the consent URL returned by get_connection_urls, then call retry_failed_post.

"Post stays as DRAFT after I asked to schedule it"

create_post only creates a draft. Your assistant must follow up with schedule_post(post_id, scheduled_for) or publish_post(post_id). The companion Antwork Poster Skill (github.com/iker-gonzalez/antwork-skills) codifies this two-step so Claude doesn't skip it.

The wrong account got the post (multiple accounts per platform)

When a workspace has more than one connected account on a platform (e.g. two LinkedIn pages), Antwork picks the first healthy account by default. To target a specific account, pass its accountId in the post's platforms array rather than the platform name. The Antwork Poster Skill also prompts the assistant to confirm before publishing.

Media upload silently drops the file

attach_media requires a Firebase Storage URL — pass external URLs through upload_media first so they get staged into your workspace storage. The publisher refuses non-storage URLs.

Rate limits

Antwork applies a soft rate limit of 60 tool calls / minute / workspace. Hitting the limit returns HTTP 429 with a Retry-After header. Publish operations are not rate-limited at the MCP layer but downstream platforms have their own limits (LinkedIn ~150 posts/day, X ~300 posts/3h).

Origin blocked / CORS error in browser-based clients

The server validates the Origin header on Streamable HTTP per the MCP spec. Allowed origins are first-party Antwork hosts, claude.ai, claude.com, console.anthropic.com, and localhost. If you're hosting a custom client on a different origin, contact us to whitelist it.

• ›Tool calls: 60 / minute / workspace (soft limit; HTTP 429 with Retry-After).
• ›Media upload: 25 MB per file. Images, video (mp4/mov/webm), PDFs accepted.
• ›Posts: no Antwork-side cap; downstream platforms enforce their own (LinkedIn ~150/day, X ~300/3h, others vary).
• ›Scheduled posts: 1 year max horizon.
• ›OAuth tokens: access token 5 min TTL, refresh token revocable any time.

Hard quotas (image generation, voice-profile refresh, brand extraction) are tied to plan tier. Pro and Business plans get higher monthly allowances; see pricing for current values.

v1.0 — May 2026

Initial public release alongside Anthropic connector directory submission.

• ›35 MCP tools across identity, social accounts, voice profiles, posts, publishing, analytics, and media.
• ›Five MCP Apps: posts-table, post-preview, calendar, upload-zone, connections-panel.
• ›OAuth 2.1 with PKCE, dynamic client registration (RFC 7591), token revocation (RFC 7009).
• ›Per-platform accountResults returned from publish_post and get_post so multi-account workspaces show the correct publishing account in UI.