MCP Server

The Middleware MCP Server is a remote Model Context Protocol (MCP) server that lets AI assistants and agentic IDEs, such as Claude Code, Claude Desktop, Cursor, and VS Code, work with your Middleware observability data using natural language.

The endpoint#

Instead of switching between your editor and the Middleware UI, ask your assistant to “list my dashboards”, “show me errors from the last hour”, or “query CPU usage for the checkout service” — it calls Middleware for you and returns the result. The server is hosted by Middleware and available at a single endpoint:

1https://mcp.middleware.io/mcp

You point a supported client at this URL, sign in to your Middleware account once in the browser, and your assistant gains access to the Middleware tools.

Connect a client#

Select your client to jump to its setup steps, or see Setup for the prerequisites and how sign-in works.

Claude CodeCLIConnect the Claude Code CLI to Middleware.Claude DesktopDesktopAdd Middleware as a custom connector in Claude Desktop.CursorIDEAdd Middleware to the Cursor IDE.VS CodeIDEUse Middleware with GitHub Copilot in VS Code.WindsurfIDEAdd Middleware to the Windsurf editor.AntigravityIDEAdd Middleware to the Antigravity editor.Gemini CLICLIConnect Middleware to the Gemini CLI.Codex CLICLIConnect Middleware to the OpenAI Codex CLI.Other clientsAnyConnect any other MCP client to Middleware.

How it works#

In MCP, the server exposes a set of tools. Each tool has a name and a typed input schema, so the assistant's calls stay structured and predictable.

1. Your assistantCalls a tool by name, with typed inputs
2. Middleware MCP ServerChecks your access token and routes to the project you chose
3. Middleware APIRuns the matching call against your data
4. Your assistantGets the result — only ever your own data

When your client invokes a tool, the server authenticates the request with your access token, routes it to the project you selected during sign-in, calls the matching Middleware API, and returns the result. You only ever see your own data.

What your assistant can do#

The server exposes 24 tools across six areas.

Dashboards — list, read, create, update, delete, clone, and favorite dashboards.

Widgets — list, create, update, delete, fetch data, and adjust layout.

Metrics and resources — discover metrics and resources, and run queries for logs, metrics, and traces.

Error tracking — list errors and incidents and read their details.

Logs — group logs into patterns and pull samples over a time window, with deep links into Middleware.

RUM — list Real User Monitoring applications, sessions, and frontend errors, with session-replay deep links.

Next steps#

Where to go from here.

SetupThe prerequisites and how sign-in works, then the setup steps for each client.MCP ToolsThe full list of tools, with each tool's inputs and behaviour.