Skip to content

Admin Console Concepts

This guide introduces each major section of the Gateway Admin UI and how it connects to the Model Context Protocol (MCP).

🆕 Setting up a new MCP Server to federate to the gateway

🔌 How do I expose an MCP server over SSE?

To federate a new MCP Server to your gateway, it must run over Server-Sent Events (SSE) so the gateway can communicate with it.

Use the built-in translate bridge to wrap any stdio-only MCP server and expose it over SSE:

python3 -m mcpgateway.translate --stdio "uvx mcp-server-git" --expose-sse --port 8001
python3 -m mcpgateway.translate --stdio "uvx mcp_server_time -- --local-timezone=Europe/Dublin" --expose-sse --port 8002

✅ Important: The gateway must be able to reach the MCP server's network address.

If you're running services inside Docker (or other containerized environments), ensure networking is configured properly: - Use host networking when needed. - Expose ports to the host machine. - Make sure internal container IPs are reachable from the gateway.

📦 Virtual Servers

A virtual server is a logical wrapper that combines selected tools, resources, and prompts under one context-specific endpoint.

🔗 What are Virtual Servers?
  • A Virtual Server defines a project-specific toolset.
  • Each one is backed by a real SSE or STDIO interface.
  • You can activate/deactivate, view metrics, and invoke tools from this server.

🛠 Global Tools

Tools are remote functions that an LLM can invoke, either via MCP or REST. Think of them like typed APIs with schemas and optional auth.

⚙️ What do Tools represent?
  • Integration Types: MCP, REST
  • Request Types: STDIO, SSE, GET, POST, etc.
  • Input Schema: JSON Schema defines valid input.
  • Supports Basic Auth, Bearer, or Custom headers.

📁 Global Resources

Resources expose read-only data like files, database rows, logs, or screenshots. LLMs can read this content through a URI.

📖 How do Resources work?
  • Text and Binary data supported.
  • Exposed via unique URI (file:///, db://, etc.).
  • Resources can be listed, templated, or subscribed to.

🧾 Global Prompts

Prompts are reusable message templates with arguments. They define system prompts, user instructions, or chainable inputs.

🗒 What's in a Prompt?
  • Each prompt has a name, template, and arguments.
  • Arguments are defined with name, description, and required status.
  • Used to enforce consistency across tool use or system messaging.

🌐 Gateways (MCP Servers)

Gateways are other MCP-compatible servers. When registered, their tools/resources/prompts become usable locally.

🌉 What is a federated Gateway?
  • Syncs public tools from a remote MCP server.
  • Peer tools show up in your catalog with gateway_id.
  • Can be toggled active/inactive.

📂 Roots

Roots define base folders for file-based resources. They control what files MCP clients can access from your local system.

📁 What are Roots used for?
  • Restrict access to specific folders (file:///workspace)
  • Prevent tools from referencing outside their sandbox.
  • Deleting a root invalidates its associated resources.

📈 Metrics

Track tool calls, resource reads, prompt renders, and overall usage in one place.

📊 What does the Metrics tab show?
  • Overall executions by server/tool/prompt.
  • Latency, failure rate, and hot paths.
  • Top tools, resources, prompts, and servers.

🧪 Version & Diagnostics

The /version endpoint returns structured JSON diagnostics including system info, DB/Redis health, and Git SHA.

🩺 What does the Version panel include?
  • MCP protocol version and server metadata.
  • Live system metrics (CPU, memory).
  • Environment checks and service readiness.

📚 Learn More