Skip to main content

Overview

MCP (Model Context Protocol) servers extend your agents’ capabilities by connecting them to external tools and services. You register remote MCP servers by providing their endpoint URL and authentication headers, then grant specific agents access to use their tools.
Only remote-hosted (HTTP) MCP servers are supported at the moment. The platform connects to your server’s endpoint URL and handles credential encryption, health monitoring, and tool discovery for you.

Registering MCP Servers

Register a new MCP server with POST /v1/mcp/servers. Provide the server’s endpoint URL and any authentication headers:
HTTP servers also support OAuth-based authentication. When configured, the platform handles token refresh and credential management automatically.
HTTP headers containing sensitive values are encrypted at rest and are never returned in API responses after creation.

Listing Servers

Retrieve your registered MCP servers with GET /v1/mcp/servers:

Getting Server Details

Retrieve a specific server’s configuration and access grants with GET /v1/mcp/servers/:id:
The response includes the server’s configuration and a list of agents that have been granted access.

Updating Servers

Update a server’s configuration with PUT /v1/mcp/servers/:id:
You can update the name, description, URL, headers, configuration, or status.

Deleting Servers

Remove a registered MCP server with DELETE /v1/mcp/servers/:id:
Deletion fails if any agents still have active access grants for this server. Revoke all agent access before deleting.

Health Checks

Check the connectivity and responsiveness of a server with GET /v1/mcp/servers/:id/health:
The response includes the server’s status and response latency, helping you diagnose connectivity issues.

Tool Refresh

Query a server’s tools/list endpoint to update the set of available tools with POST /v1/mcp/servers/:id/tools/refresh:
Run this after updating a server’s configuration or deploying a new version of the underlying MCP server to ensure the platform has an up-to-date list of available tools.

Access Management

Control which agents can use an MCP server by managing access grants.

Grant Access

Grant an agent access to a server with POST /v1/mcp/servers/:id/access:
To restrict an agent to specific tools:
Use selected access mode with allowed_tools in production to follow the principle of least privilege. Only grant agents access to the specific tools they need.

List Access Grants

See which agents have access to a server with GET /v1/mcp/servers/:id/access:

Revoke Access

Remove an agent’s access with DELETE /v1/mcp/servers/:id/access/:agentId:

Example: GitHub Integration

This example walks through registering a GitHub MCP server and granting an agent access to it. Step 1: Register the GitHub MCP server:
Step 2: Grant your agent access with selected tools:
Step 3: Verify connectivity:
Your agent can now use the GitHub tools in its conversations.

Best Practices

  • Encrypt credentials properly. Always pass secrets through headers. Never hardcode credentials in URLs.
  • Use selected access mode in production. Restrict agents to only the tools they need rather than granting blanket all access.
  • Run health checks after configuration changes. Verify server connectivity whenever you update credentials or URLs.
  • Refresh tools after server updates. When the underlying MCP server is updated with new tools, call the tool refresh endpoint to update the platform’s tool registry.
  • Revoke access before deleting servers. Clean up all agent access grants before removing a server to avoid errors.

Next Steps

Agent Configuration

Configure agents and connect them to MCP servers.

Agent Messaging

Send messages and manage conversations.

Filesystem Management

Manage files in agent workspaces.