MCP Integration

FanDesk implements the Model Context Protocol (MCP) — the open standard adopted by Anthropic, OpenAI, Google, and Microsoft — giving any MCP-compatible AI agent direct, authenticated access to your entire FanDesk workspace through 250+ tools across every module.

What is MCP?

MCP (Model Context Protocol) is an open standard backed by Anthropic and the Linux Foundation. It defines a universal interface for AI clients to call tools, read resources, and take actions in external applications — without writing custom API integration code.

FanDesk is a first-mover on MCP with 250+ tools exposed, covering every module: tasks, email, calendar, files, chat, pages, incidents, contacts, knowledge graph, pulse meetings, forms, automations, and more.

What this means for your AI agents: Your AI agent gets direct access to your entire workspace. Instead of writing code to interact with APIs, it calls FanDesk tools directly. Fewer tokens, zero broken code, instant execution. An agent that used to need 10 minutes and 50,000 tokens to build and debug an API integration can now accomplish the same task in seconds with a single tool call.

Why MCP + FanDesk is Different

Most MCP servers expose a handful of read-only tools for a single application. FanDesk exposes 250+ read/write tools across your entire work stack — the equivalent of giving your AI agent accounts in Slack, Jira, Notion, Gmail, Google Calendar, Google Drive, PagerDuty, and your CRM, all from one authenticated connection.

Because FanDesk stores all your data natively with a shared knowledge graph, your AI agent can do things that are impossible with fragmented tools:

• Read an email, create a task from it, assign it to a project, and post a summary in the right chat channel — all in one conversation turn
• Find a meeting transcript, convert it to a page, extract action items into tasks, and send a follow-up email — zero switching between tools
• Search semantic knowledge across all your data (emails, tasks, pages, meetings) in one query

Supported Clients

Quick Start

Step 1: Generate an API Key

Go to Settings > Integrations and click Generate Key in the API Keys section. Give your key a descriptive name (e.g., "Claude Code" or "Cursor").

Important: Copy the key immediately. For security, FanDesk stores only a hash — you cannot retrieve the raw key after closing the dialog.

Step 2: Connect Claude Code

Run this in your terminal from any project directory:

claude mcp add fandesk \
  --transport streamable-http \
  --url https://app.fandesk.ai/mcp \
  --header "Authorization: Bearer fdk_your_key_here"

That's it. Claude Code will now have access to all 250+ FanDesk tools in every conversation. Verify the connection with:

claude mcp list

Step 3: Connect Claude Desktop

Add to your Claude Desktop config file (claude_desktop_config.json):

{
  "mcpServers": {
    "fandesk": {
      "url": "https://app.fandesk.ai/mcp",
      "headers": {
        "Authorization": "Bearer fdk_your_key_here"
      }
    }
  }
}

Restart Claude Desktop after saving. The FanDesk tools will appear in Claude's tool list automatically.

Step 4: Connect Cursor or Windsurf

In Cursor, go to Settings > MCP and add a new server:

{
  "fandesk": {
    "url": "https://app.fandesk.ai/mcp",
    "headers": {
      "Authorization": "Bearer fdk_your_key_here"
    }
  }
}

Example: What Agents Can Do

Once connected, your agent works across your entire workspace in natural language:

"List my open tasks in the Backend project, find the most urgent one,
 check if there is an email thread about it, and post a message to
 #engineering with a status update."

This single prompt triggers: list_tasks then search_emails then send_message — three tools, one conversation turn, zero custom code.

Sales outbound automation: Use project tools to create a sales pipeline, mail tools to send personalized outreach, contact tools to track interactions, and calendar tools to schedule follow-ups — all from one conversation.

Code automation: Use project tools to break down features into tasks with full context. Your AI coding agent reads one task, understands scope and dependencies, executes, and moves to the next. Structured work means radical token efficiency.

Incident response: When an alert fires, create an incident, page the on-call engineer via chat, and link the relevant task — then generate a postmortem once it is resolved.

How Permissions Work

Each API key inherits exactly the permissions of the user who created it. Your AI agent cannot access anything you could not access yourself:

No admin privileges are granted regardless of tool calls attempted. All permission checks run server-side on every request.

Security

• API keys are hashed with bcrypt — FanDesk never stores the raw key
• Keys can be revoked instantly from Settings > Integrations
• Every tool call is logged and attributed to your user account
• All traffic uses TLS — no plaintext API keys in transit
• Rate limiting and abuse detection are enforced per key
• Keys are scoped strictly to the creating user's permissions

Transport Details

FanDesk uses the Streamable HTTP MCP transport (the modern standard). The endpoint is:

POST https://app.fandesk.ai/mcp
Authorization: Bearer fdk_your_key_here
Content-Type: application/json

All requests are authenticated. The server supports streaming responses for long-running tool calls.

Frequently Asked Questions

Can I connect multiple AI clients to the same FanDesk account? Yes. Generate a separate API key for each client (e.g., one for Claude Code, one for Cursor). Each key is independent and can be revoked separately.

Do tool calls count against my FanDesk plan? MCP tool calls are standard API operations — they count the same as actions taken in the app. Heavy automation use may require a higher plan tier.

Can I use MCP with a self-hosted FanDesk instance? Yes. Replace app.fandesk.ai with your instance URL. All 250+ tools are available on self-hosted instances.

What happens if a tool call fails? The MCP server returns a structured error response. Well-behaved MCP clients (Claude Code, Claude Desktop) will surface the error message to the user or agent automatically.