← Back to Help Center

Connect AI Assistants via MCP

Step-by-step guide to connect Claude, ChatGPT, Perplexity, Manus AI, and other AI assistants to FeedMansion.

Note: The Help Center is under construction. Articles may be incomplete, outdated, or still evolving.

GitHub Repo: For platform-specific setup guides and example prompts, see github.com/makerprism/feedmansion-ai-agents

What is MCP?

The Model Context Protocol (MCP) is an open standard that lets AI assistants connect to external tools and services. FeedMansion provides a built-in MCP server so your AI assistant can help you manage content. Draft posts, organize queues, and more - all from within your AI chat.

No installation or self-hosting required. FeedMansion's MCP server runs remotely at https://feedmansion.com/mcp.

Prerequisites

Before connecting an AI assistant, you need an Agent Token:

1. Go to Settings > API Tokens in FeedMansion
2. Choose the Agent token type
3. Optionally restrict the token to specific presences
4. Click Create Token and copy the token (it starts with fm_pat_)

Agent tokens include a supervision gate: your AI assistant can create and organize content, but nothing gets published without your explicit approval in the FeedMansion UI.

Claude

Claude.ai (Web / Desktop)

1. Open Claude > Settings > Integrations (or Connectors)
2. Click Add MCP Server (or Add Custom Connector)
3. Enter the server URL: https://feedmansion.com/mcp
4. For authentication, add a custom header:
   • Header name: Authorization
   • Header value: Bearer fm_pat_YOUR_TOKEN_HERE
5. Save. Claude will discover the available FeedMansion tools automatically.

Claude Code / Claude Desktop (JSON config)

Add to your MCP configuration file:

{
  "mcpServers": {
    "feedmansion": {
      "type": "streamable-http",
      "url": "https://feedmansion.com/mcp",
      "headers": {
        "Authorization": "Bearer fm_pat_YOUR_TOKEN_HERE"
      }
    }
  }
}

ChatGPT

1. Open ChatGPT > Settings > MCP
2. Click Add MCP Server
3. Enter the server URL: https://feedmansion.com/mcp
4. Set the authentication header: Authorization: Bearer fm_pat_YOUR_TOKEN_HERE
5. Save. ChatGPT will discover the available tools.

Perplexity

1. Open Perplexity > Settings > MCP
2. Click Add Server
3. Enter the URL: https://feedmansion.com/mcp
4. Add your token as a Bearer authorization header
5. Save and test the connection

Manus AI

1. Open Manus AI > Settings > MCP Servers
2. Add a new server with URL: https://feedmansion.com/mcp
3. Configure authentication with your Agent Token as a Bearer token
4. Save. Manus will detect the available FeedMansion tools.

OpenClaw

Self-hosted option. OpenClaw runs on your own infrastructure, giving you full control and privacy. Best suited for users comfortable with server administration and security configuration.

Security note: Properly securing an OpenClaw deployment (authentication, network exposure, token management, updates) is nontrivial. If misconfigured, it could expose your FeedMansion account or other connected services.

For a simpler and more secure setup without self-hosting, try Claude or Cursor instead.

OpenClaw is a personal AI assistant platform that runs on your own infrastructure.

Using the FeedMansion Skill

1. Copy the skill to your OpenClaw workspace:

# Clone the AI agents repo
git clone https://github.com/makerprism/feedmansion-ai-agents.git

# Copy the skill
cp feedmansion-ai-agents/platforms/openclaw/SKILL.md ~/.openclaw/workspace/skills/feedmansion/SKILL.md

1. Configure your PAT in OpenClaw (~/.openclaw/openclaw.json):

{
  "env": {
    "FEEDMANSION_PAT": "fm_pat_YOUR_TOKEN_HERE"
  }
}

1. Restart OpenClaw:

openclaw gateway restart

Now you can use FeedMansion tools directly in OpenClaw! Try:

• "List my FeedMansion presences"
• "Generate a post from https://example.com/article"
• "Create a draft for presence 'abc123' with text 'Hello world!'"

For full documentation, see the OpenClaw platform guide.

Cursor / Other MCP Clients

Any MCP client that supports Streamable HTTP transport can connect using this JSON config:

Replace `fm_pat_YOUR_TOKEN_HERE` with your actual Agent Token.

## Tool Approval

All FeedMansion tools are safe to auto-approve. None of them publish content directly. The server-side supervision gate ensures nothing goes live without your review in the FeedMansion web UI. You can allow your AI client to call all FeedMansion tools without asking for confirmation each time.

## What Can AI Assistants Do?

Once connected, your AI assistant has access to these tools:

**Discovery**

* **list\_presences** \- See all your presences (brands/projects)
* **get\_presence** \- Get details about a presence, including connected accounts, queues, and ghost writers
* **list\_drafts** \- Browse drafts, optionally filtered by status
* **get\_draft** \- Read a specific draft with its details
* **get\_schedule** \- See what's scheduled, queued, and recently published for a date range

**Create drafts**

* **create\_draft** \- Create a draft using AI in your brand voice. Optionally generate from a URL or attach media. Ghost writers and custom prompts are optional.
* **batch\_create\_drafts** \- Create up to 10 drafts at once

**Brand voice + media**

* **get\_brand\_voice** \- Retrieve your complete brand voice for external content creation
* **edit\_draft** \- Edit an existing draft's text or title
* **add\_draft\_media** \- Attach an image or video to a draft (from URL, presigned upload, or inline base64)
* **remove\_draft\_media** \- Remove an attached media item from a draft

**Schedule drafts**

* **schedule\_draft** \- Approve a draft and add it to a publishing queue in one step
* **batch\_schedule\_drafts** \- Schedule multiple drafts at once with shared target accounts and queue

**Feedback**

* **submit\_bug\_report** \- Report a bug directly from your AI chat
* **submit\_feature\_request** \- Suggest a new feature or improvement

## Safety: The Supervision Gate

Agent tokens enforce a **supervision gate**. Your AI assistant can prepare content, but critical actions require your approval:

* **Creating drafts**: Allowed. Drafts are saved with status "draft".
* **Approving content**: The AI submits content for **your review**. It moves to "pending review" status, not directly to publishing.
* **Enqueuing content**: The AI saves an enqueue _intent_. You confirm it in the FeedMansion UI.
* **Publishing**: Not available to AI assistants. Only humans can publish.

This means your AI assistant is a powerful content co-pilot, but you always have the final say before anything goes live.