MCP Servers & Transports

AVAILABLE

Connect external Model Context Protocol (MCP) servers to seamlessly expose remote filesystems, databases, APIs, and tools to your AI applications.

  • Portal Location: Projects > [Your Project] > AI & Capabilities > MCP Servers (/projects/[slug]/mcp-servers)
  • API Endpoint: /v1/mcp/servers

What is MCP (Model Context Protocol)? #

MCP is an open standard developed to connect AI models with external data sources and tools.

Naagmani provides a built-in MCP Gateway that manages:

  1. Connection Lifecycle: Maintains long-lived transports (stdio subprocesses or SSE HTTP streams) to external MCP servers.
  2. Dynamic Tool Discovery: Automatically discovers tools, resources, and prompts exposed by the remote MCP server.
  3. Governance & Approval Policies: Applies fine-grained MCP Tool Policies before executing sensitive actions.
Rendering diagram...

Supported Transport Protocols #

TransportConnection TargetConfiguration Fields
stdioLocal binary or command on gateway host.Command (npx -y @modelcontextprotocol/server-postgres), args, environment variables.
sseRemote HTTP/HTTPS Server-Sent Events stream.Remote URL (https://mcp.internal.company.com/sse), Bearer authorization headers.

Connecting an MCP Server in Developer Portal #

  1. In your project workspace, click AI & Capabilities > MCP Servers.
  2. Click Add MCP Server.
  3. Configure:
    • Server Name: e.g., GitHub MCP Connector.
    • Transport Type: Select stdio or sse.
    • Command / URL: Endpoint or executable command.
    • Environment Variables: e.g. GITHUB_PERSONAL_ACCESS_TOKEN.
  4. Click Connect & Probe. Naagmani executes the MCP handshake and lists all available remote tools.