MindThread

MindThread MCP: the Threads scheduler built for AI agents

Let Claude, Cursor or any MCP client write, schedule and publish Threads posts, review drafted replies and read insights for your accounts, through Meta's official Threads API. 40+ tools, a free sandbox key, and permission controls so the agent asks before it posts.

claude mcp add --transport http mindthread https://mindthread.tw/api/mcp --header "Authorization: Bearer mt_live_your_key"

繁體中文說明在下方

What your agent can do

• Write: generate 1 to 5 drafts in the account's voice (persona, brand material, content formulas), or threads of 2 to 10 parts.
• Schedule and publish: text, single image, carousel (2 to 10 images) or video, now or at an exact time; or drop posts into a queue that publishes at the account's own time slots.
• Stay on-brand in code: content rules (length, banned phrases, no hashtags, only numbers from your fact list) are enforced by the server, not just suggested in a prompt.
• Reply where it matters: cloud patrol searches Threads for your keywords and drafts replies; your agent lists the drafts and approves or rejects each one.
• Measure: views, likes, replies and followers per account, plus the results of patrol replies.
• Research and plan (MindThread API 1.22.0): rank the best posts of an account you connected (get_top_posts; the Threads API only returns insights for the authorized user's own posts, so other people's accounts cannot be read) and keep brand, content and content-map documents on MindThread (save_brand_doc, get_brand_doc, last 20 versions kept).
• Stay in control: every key has per-permission allow, ask or deny. With ask, the post waits until you confirm it on mindthread.tw, and the agent gets a link to hand you instead of posting.

Three-minute setup

1. Get an API key. Sign up at mindthread.tw, connect a Threads account, then open Settings → API keys and create a key. Start with a test key (mt_test_…): reads show your real data, writes are simulated and nothing reaches Threads. Switch to a live key (mt_live_…) when you are ready. Every plan includes API access (October 2026: free plan 50 calls a day per live key; test keys share a 500-a-day sandbox). Current limits: mindthread.tw/developers.

2. Add MindThread to your client (pick yours below).

3. Ask for your first scheduled post:

List my MindThread accounts. Then write a short post about why I started my business, show it to me, and schedule it on my account for tomorrow at 9:00 Taipei time.

The agent calls list_accounts, generate_posts, then publish with publish_at. Check the result with "show my scheduled posts" (list_schedules).

Client setup

There are two ways to connect, and both reach the same server:

• Remote (recommended): point the client at https://mindthread.tw/api/mcp. Nothing to install.
• Local stdio: run this package with npx -y mindthread-mcp. Use it for clients that only start servers with a command. It forwards every message to the same endpoint, so the tools are always identical.

Claude Code

# Remote, with an API key

claude mcp add --transport http mindthread https://mindthread.tw/api/mcp --header "Authorization: Bearer mt_live_your_key"

Remote, signing in with your MindThread account instead of a key (run /mcp in Claude Code to log in)

claude mcp add --transport http mindthread https://mindthread.tw/api/mcp

Local stdio

claude mcp add mindthread --env MINDTHREAD_API_KEY=mt_live_your_key -- npx -y mindthread-mcp

Claude Desktop

Edit claude_desktop_config.json (Settings → Developer → Edit Config):

{

"mcpServers": { "mindthread": { "command": "npx", "args": ["-y", "mindthread-mcp"], "env": { "MINDTHREAD_API_KEY": "mt_live_your_key" } } } }

On Windows, if Claude cannot find npx, use "command": "cmd" and "args": ["/c", "npx", "-y", "mindthread-mcp"]. Without editing files: Settings → Connectors → Add custom connector, URL https://mindthread.tw/api/mcp, then sign in with your MindThread account (on Claude plans that offer custom connectors).

Cursor

~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project):

{

"mcpServers": { "mindthread": { "url": "https://mindthread.tw/api/mcp", "headers": { "Authorization": "Bearer mt_live_your_key" } } } }

Or the stdio form: "command": "npx", "args": ["-y", "mindthread-mcp"], "env": { "MINDTHREAD_API_KEY": "mt_live_your_key" }.

VS Code (GitHub Copilot agent mode)

.vscode/mcp.json, which asks for the key once and stores it securely:

{

"inputs": [ { "type": "promptString", "id": "mindthread-key", "description": "MindThread API key", "password": true } ], "servers": { "mindthread": { "type": "http", "url": "https://mindthread.tw/api/mcp", "headers": { "Authorization": "Bearer ${input:mindthread-key}" } } } }

Any other MCP client

• Streamable HTTP: https://mindthread.tw/api/mcp, header Authorization: Bearer (also accepted as X-API-Key).
• stdio: command npx, args -y mindthread-mcp, env MINDTHREAD_API_KEY.

Without a key, initialize, tools/list, get_playbook and list_formulas still work, so you can look around first.

Example prompts

• "What can MindThread do? Read the playbook first." (get_playbook)
• "Write 3 posts about saving for a first home in @mybrand's voice. Don't post anything yet." (generate_posts)
• "Queue posts 1 and 3; schedule post 2 for Friday 20:00." (add_library, publish)
• "Any drafted replies waiting for me? Show them one by one." (list_pending_replies, approve_pending_reply, reject_pending_reply)
• "How did @mybrand do in the last 7 days compared with the 7 days before?" (analytics_insights)
• "Rank the 20 best posts of @mybrand by engagement and summarize my core ideas and voice into a content document." (get_top_posts, save_brand_doc)
• Three Traditional Chinese prompt templates (best posts, brand interview, content map): mindthread-cli README and mindthread.tw/developers#prompt-templates.
• "Set cloud patrol keywords to 'first home', 'mortgage', 'rent vs buy', at most 10 replies a day, and hold every reply for my review." (set_cloud_patrol)

Tools

The set any client sees before signing in, generated from the live server with npm run tools-table. The server is the source of truth: when it adds a tool, this package exposes it without an update.

| reject_pending_reply | Reject a drafted reply | write | Rejects one drafted cloud patrol reply by id with an optional note.