Model Context Protocol (MCP)

Integrate MCP servers with Cascade to access custom tools like GitHub, databases, and APIs. Configure stdio, HTTP, and SSE transports with admin controls for Teams.

MCP (Model Context Protocol) is a protocol that enables LLMs to access custom tools and services. An MCP client (Cascade, in this case) can make requests to MCP servers to access tools that they provide. Cascade now natively integrates with MCP, allowing you to bring your own selection of MCP servers for Cascade to use. See the official MCP docs for more information.

Enterprise users must manually turn this on via settings

Adding a new MCP

New MCPs can be added from the MCP Marketplace, which you access by clicking on the MCPs icon in the top right menu in the Cascade panel, or from the Windsurf Settings > Cascade > MCP Servers section.

If you cannot find your desired MCP, you can add it manually by editing the raw mcp_config.json file.

Official MCPs will show up with a blue checkmark, indicating that they are made by the parent service company.

When you click on a MCP, simply click Install to expose the server and its tools to Cascade.

Windsurf supports three transport types for MCP servers: stdio, Streamable HTTP, and SSE.

Windsurf also supports OAuth for each transport type.

For http servers, the URL should reflect that of the endpoint and resemble https://<your-server-url>/mcp.

Configuring MCP tools

Each MCP has a certain number of tools it has access to. Cascade has a limit of 100 total tools that it has access to at any given time.

On each MCP settings page, you can toggle the tools that you wish to enable. To open settings for a MCP, click on the MCPs icon in the top right menu in the Cascade panel, and click on the desired MCP.

mcp_config.json

The ~/.codeium/windsurf/mcp_config.json file is a JSON file that contains a list of servers that Cascade can connect to.

Here’s an example configuration, which sets up a single server for GitHub:

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "<YOUR_PERSONAL_ACCESS_TOKEN>"
      }
    }
  }
}

Be sure to provide the required arguments and environment variables for the servers that you want to use.

See the official MCP server reference repository or OpenTools for some example servers.

Remote HTTP MCPs

It’s important to note that for remote HTTP MCPs, the configuration is slightly different and requires a serverUrl or url field.

Here’s an example configuration for an HTTP server:

{
  "mcpServers": {
    "remote-http-mcp": {
      "serverUrl": "<your-server-url>/mcp",
      "headers": {
        "API_KEY": "value"
      }
    }
  }
}

Config Interpolation

The ~/.codeium/windsurf/mcp_config.json file handles interpolation of environment variables in these fields: command, args, env, serverUrl, url, and headers.

Here’s an example configuration, which uses an AUTH_TOKEN environment variable in headers.

{
  "mcpServers": {
    "remote-http-mcp": {
      "serverUrl": "<your-server-url>/mcp",
      "headers": {
        "API_KEY": "Bearer ${env:AUTH_TOKEN}"
      }
    }
  }
}

Admin Controls (Teams & Enterprises)

Team admins can toggle MCP access for their team, as well as whitelist approved MCP servers for their team to use:

MCP Team Settings

Configurable MCP settings for your team.

The above link will only work if you have admin privileges for your team.

By default, users within a team will be able to configure their own MCP servers. However, once you whitelist even a single MCP server, all non-whitelisted servers will be blocked for your team.

How Server Matching Works

When you whitelist an MCP server, the system uses regex pattern matching with the following rules:

• Full String Matching: All patterns are automatically anchored (wrapped with ^(?:pattern)$) to prevent partial matches
• Command Field: Must match exactly or according to your regex pattern
• Arguments Array: Each argument is matched individually against its corresponding pattern
• Array Length: The number of arguments must match exactly between whitelist and user config
• Special Characters: Characters like $, ., [, ], (, ) have special regex meaning and should be escaped with \ if you want literal matching

Configuration Options

Option 1: Plugin Store Default (Recommended)

Leave the Server Config (JSON) field empty to allow the default configuration from the Windsurf MCP Plugin Store.

Admin Whitelist Configuration:

• Server ID: github-mcp-server
• Server Config (JSON): (leave empty)

{}

Matching User Config (mcp_config.json):

{
  "mcpServers": {
    "github-mcp-server": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "GITHUB_PERSONAL_ACCESS_TOKEN",
        "ghcr.io/github/github-mcp-server"
      ],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_your_token_here"
      }
    }
  }
}

This allows users to install the GitHub MCP server with any valid configuration, as long as the server ID matches the plugin store entry.

Option 2: Exact Match Configuration

Provide the exact configuration that users must use. Users must match this configuration exactly.

Admin Whitelist Configuration:

• Server ID: github-mcp-server
• Server Config (JSON):

{
  "command": "docker",
  "args": [
    "run",
    "-i",
    "--rm",
    "-e",
    "GITHUB_PERSONAL_ACCESS_TOKEN",
    "ghcr.io/github/github-mcp-server"
  ],
  "env": {
    "GITHUB_PERSONAL_ACCESS_TOKEN": ""
  }
}

Matching User Config (mcp_config.json):

{
  "mcpServers": {
    "github-mcp-server": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "GITHUB_PERSONAL_ACCESS_TOKEN",
        "ghcr.io/github/github-mcp-server"
      ],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_your_token_here"
      }
    }
  }
}

Users must use this exact configuration - any deviation in command or args will be blocked. The env section can have different values.

Option 3: Flexible Regex Patterns

Use regex patterns to allow variations in user configurations while maintaining security controls.

Admin Whitelist Configuration:

• Server ID: python-mcp-server
• Server Config (JSON):

{
  "command": "python3",
  "args": ["/.*\\.py", "--port", "[0-9]+"]
}

Matching User Config (mcp_config.json):

{
  "mcpServers": {
    "python-mcp-server": {
      "command": "python3",
      "args": ["/home/user/my_server.py", "--port", "8080"],
      "env": {
        "PYTHONPATH": "/home/user/mcp"
      }
    }
  }
}

This example allows users flexibility while maintaining security:

• The regex /.*\\.py matches any Python file path like /home/user/my_server.py
• The regex [0-9]+ matches any numeric port like 8080 or 3000
• Users can customize file paths and ports while admins ensure only Python scripts are executed

Common Regex Patterns

Pattern	Matches	Example
.*	Any string	/home/user/script.py
[0-9]+	Any number	8080, 3000
[a-zA-Z0-9_]+	Alphanumeric + underscore	api_key_123
\\$HOME	Literal $HOME	$HOME (not expanded)
\\.py	Literal .py	script.py
\\[cli\\]	Literal [cli]	mcp[cli]

Notes

Admin Configuration Guidelines

• Environment Variables: The env section is not regex-matched and can be configured freely by users
• Disabled Tools: The disabledTools array is handled separately and not part of whitelist matching
• Case Sensitivity: All matching is case-sensitive
• Error Handling: Invalid regex patterns will be logged and result in access denial
• Testing: Test your regex patterns carefully - overly restrictive patterns may block legitimate use cases

Troubleshooting

If users report that their MCP servers aren’t working after whitelisting:

1. Check Exact Matching: Ensure the whitelist pattern exactly matches the user’s configuration
2. Verify Regex Escaping: Special characters may need escaping (e.g., \. for literal dots)
3. Review Logs: Invalid regex patterns are logged with warnings
4. Test Patterns: Use a regex tester to verify your patterns work as expected

Remember: Once you whitelist any server, all other servers are automatically blocked for your team members.

General Information

• Since MCP tool calls can invoke code written by arbitrary server implementers, we do not assume liability for MCP tool call failures. To reiterate:
• We currently support an MCP server’s tools, resources, and prompts.