Skip to main content
Connect MCP (Model Context Protocol) servers to extend your AI agent with powerful integrations. MCP is a standard protocol that enables AI agents to interact with external tools and services.

What is MCP?

Model Context Protocol (MCP) is an open standard for connecting AI models to external data sources and tools. MCP servers provide a standardized way for your agent to:
  • Access external services
  • Execute tools and actions
  • Retrieve contextual information

Connecting an MCP Server

Step 1: Get Server Details

You’ll need:
  • Server URL — The HTTP/HTTPS endpoint of the MCP server
  • Transport Protocol — How to communicate (Streamable HTTP or SSE)
  • Headers (optional) — Any authentication headers required

Step 2: Add the Server

  1. Go to Integration > MCP Servers
  2. Click Add Server
  3. Fill in the connection details:
FieldDescription
Server NameA friendly name (e.g., “my-api-server”)
Server URLThe MCP endpoint URL
TransportStreamable HTTP (recommended) or SSE
HeadersOptional auth headers in Key: Value format

Step 3: Test Connection

Click Test Connection to verify the server is reachable and responding correctly.

Step 4: Save

Once the connection test passes, save the server configuration.

Transport Protocols

The modern transport protocol for MCP. Use this unless you have a specific reason to use SSE.

SSE (Server-Sent Events)

Legacy transport protocol. Some older MCP servers may only support SSE.

Managing Connected Servers

View Connected Servers

All connected servers appear in a list showing:
  • Server name
  • Connection status
  • Number of available tools
  • Enable/disable toggle

Enable/Disable Servers

Toggle a server on or off without removing it. Disabled servers won’t be available to your agent.

View Available Tools

Click the expand arrow on any server to see its available tools. Each tool shows:
  • Tool name
  • Description of what it does

Edit Server

Click the edit button to modify:
  • Server URL
  • Transport protocol
  • Headers

Delete Server

Remove a server entirely by clicking the delete button.

Configuring Tools

After connecting a server, you can fine-tune which tools your agent can use:
  1. Go to Actions > MCP Servers
  2. Find your server in the list
  3. Expand to see individual tools
  4. Toggle specific tools on or off
This gives you granular control over what your agent can do.

Pre-Built Integrations

Crow includes pre-built support for popular services:

Notion

Connect your Notion workspace via Knowledge Base > Notion. This uses MCP under the hood to provide real-time access to your Notion content.

Coming Soon

  • Slack
  • GitHub
  • Google Drive

Building Custom MCP Servers

If you want to build your own MCP server to expose custom tools:
  1. Implement the MCP protocol specification
  2. Host your server at a publicly accessible URL
  3. Add the server URL in Crow
MCP servers can be built in any language that supports HTTP.

Best Practices

Use Descriptive Names

Name your servers clearly so you can identify them later:
  • Good: customer-data-api, order-management
  • Bad: server1, test

Secure Your Endpoints

  • Use HTTPS for all MCP server URLs
  • Include authentication headers when required
  • Don’t expose sensitive operations unnecessarily

Test After Connecting

Verify in the Sandbox that your agent can discover and use the MCP tools correctly.

Monitor for Changes

If your MCP server adds or removes tools, Crow will discover changes automatically when tools are fetched.

Troubleshooting

Connection Test Failed

  • Verify the server URL is correct
  • Check that the server is running and accessible
  • Ensure any required headers are included
  • Try the other transport protocol

Tools Not Appearing

  • Expand the server to trigger tool discovery
  • Check server logs for errors
  • Verify the server implements tool listing correctly

Agent Not Using MCP Tools

  • Enable the server in the MCP servers list
  • Enable specific tools in Actions
  • Test with queries that should trigger the tools

Need Help?