Cursor MCP server: there isn't one, and why

Cursor sits on the other side of the protocol. What that means if you came looking for a Cursor connector, and what to connect to Cursor instead.

Last verified 1 September 2026 · from Cursor's changelog and documentation, its GitHub organization and the MCP registry

This page is one of 90 in a directory of monitoring and developer tools. Each page checks what an AI coding agent can get out of the tool through its MCP server, the connector that lets the agent query the tool directly.

What is the Cursor MCP server, and is it usable?

Cursor is an AI code editor. Its role in the Model Context Protocol is to be the thing an agent runs inside. You configure servers in ~/.cursor/mcp.json for every project, or .cursor/mcp.json for one, and Cursor makes those tools available to the model.

Every vendor page in this guide has a Cursor configuration block for that reason. The connector belongs to the vendor whose data it exposes; Cursor is where it gets used.

There is no Cursor-shaped data an MCP server would expose. An editor's state is the file you have open, which the agent already has.

Every one of the 90 vendors in this directory gets the same four checks. Cursor passes 0 of them.

Cursor operates in code, tests and review as the thing an agent runs inside. It publishes no server of its own, by design.

The takeaway is a redirection rather than a verdict. Cursor has nothing to connect to because it is the thing you connect things to. If you arrived looking for a Cursor MCP server, the page you want is the one for whichever product holds the data you need, and its Cursor setup block.

What can an AI agent do with Cursor?

One thing goes wrong here, and it is a search rather than a server.

How to configure Cursor MCP for agents

1. 1

   Choose the scope

   .cursor/mcp.json in a project applies to that project. ~/.cursor/mcp.json applies everywhere.

2. 2

   Pick the transport deliberately

   Cursor documents stdio as local, single-user and manually authenticated, while SSE and Streamable HTTP are deployed servers that support multiple users and OAuth.

3. 3

   Interpolate rather than hardcode

   Cursor resolves variables in command, args, env, url and headers, so a secret can come from the environment instead of the file.

4. 4

   Set the team allowlist before rollout

   Cursor states that allowlisting approves a configuration; it does not distribute or install the server.

5. 5

   Use tool allowlists to narrow a server

   They restrict which tools from an approved server run automatically, and an empty list allows all of them.

Paste this into your agent’s instructions

Cursor supports more of the protocol than tools alone, so an answer can come from a resource or a prompt rather than a tool call. Naming the source keeps that visible.

When you answer inside Cursor, state:
- Which server answered, and whether the answer came from a
  tool, a resource or a prompt. Cursor supports all three.
- Whether the server was project-scoped or global, because
  the two files can define different servers.
Never state that data does not exist. State which server you
queried and what it returned.
Write your answer in ASD-STE100 Simplified Technical
English. Use short sentences with one idea in each.

Do you need the MCP server at all?

There is no Cursor server to need, so the question becomes what to connect to Cursor. Cursor reads ~/.cursor/mcp.json for all projects and .cursor/mcp.json for one, and both take the same shape as the blocks on the vendor pages here.

For code and pipelines, GitHub and CircleCI are the usual pair. For what happens after a deploy, Grafana, Datadog and Dynatrace. For your own agents, Langfuse, Arize Phoenix and LangSmith.

One practical note that applies to all of them: every server you add spends context before you ask anything, and that budget is shared. Connecting five servers of the size measured on the vendor pages here can cost a fifth of the window before the first question.

Pick the servers for the data you actually query in Cursor, and keep the list short. The cost is paid at connection, not at use.

Cursor MCP server setup

Nothing here installs a Cursor server, because there is none. This is how other vendors' servers are added to Cursor, which is what a reader on this page is usually after.

Two files decide scope: .cursor/mcp.json for one project, ~/.cursor/mcp.json for everywhere. Both use the mcpServers key.

No credential needs to sit in the file. Cursor's interpolation resolves ${env:NAME} inside command, args, env, url and headers.

{"mcpServers":{"server-name":{
  "command":"npx",
  "args":["-y","mcp-server"],
  "env":{"API_KEY":"value"}}}}

{"mcpServers":{"server-name":{
  "url":"https://api.example.com/mcp"}}}

{"mcpServers":{"server-name":{
  "url":"https://api.example.com/mcp",
  "auth":{
    "CLIENT_ID":"${env:MCP_CLIENT_ID}",
    "CLIENT_SECRET":"${env:MCP_CLIENT_SECRET}"}}}}

${env:NAME}   ${userHome}   ${workspaceFolder}
${workspaceFolderBasename}   ${pathSeparator}  ${/}

Every other client

Each block below is the configuration for one client, with the file path and the key that client expects.

Team Settings  ->  MCP Configuration
  command entries  ->  approve local stdio servers
  URL entries      ->  approve remote HTTP/SSE servers
  tool allowlists  ->  empty means all tools allowed

vscode.cursor.mcp.registerServer()

Read-only access, permissions and security

• Allowlisting is approval, not installation

  Cursor states plainly that allowlisting approves an MCP configuration. It does not distribute or install the server. An approved entry still has to be installed by the user.

• An empty tool allowlist means every tool

  Leaving the list empty allows all tools from that server to run automatically, which is the opposite of what an empty list usually implies.

• stdio authenticates manually

  Cursor's transport table marks stdio as manual auth and single user, while SSE and Streamable HTTP support OAuth and multiple users.

• Secrets belong in the environment

  Cursor's documented example uses ${env:MCP_CLIENT_ID} and ${env:MCP_CLIENT_SECRET} rather than literal values in mcp.json.

• The surface is wider than tools

  Cursor supports tools, prompts, resources, roots, elicitation and interactive apps. Roots tell a server which folders it may see. Elicitation lets a server ask the user a question mid-call. Reviewing a server therefore means reviewing more than its tool list.

Troubleshooting

You cannot find a Cursor MCP server to install

There is not one. Cursor is the client. Search instead for the vendor whose data you want, and add that server to mcp.json.

A server works in one project and not another

Check which file defines it. .cursor/mcp.json is project-scoped; ~/.cursor/mcp.json applies everywhere.

A variable appears literally instead of resolving

Interpolation applies to command, args, env, url and headers only. A variable placed in any other field is left as written.

An allowlisted server still is not available

Allowlisting approves a configuration; it does not install anything. The server still has to be configured on the machine.

Tools run automatically when you did not expect it

An empty tool allowlist allows every tool from that server. Name the tools explicitly to narrow it.

Cursor MCP server: Reference

What engineers report

There is no server to have opinions about, so this section records what the search for one actually finds.

DNS checked and the registry and organization searched on 24 August 2026. Cursor's role as a client is documented by Cursor itself.

Should you connect an agent to Cursor?

There is nothing to connect: this is the agent side of MCP. Best for connecting a small, deliberate set of vendor servers rather than looking for a Cursor one.

Strongest fit

Teams who want an editor that can call other vendors' MCP servers, which is what Cursor is for.

Main advantage

Mature client support, with per-project and global server configuration.

Main weakness

Nothing to connect to. There is no Cursor data an agent could query.

Operational risk

Every server added to Cursor spends context at connection, and the budget is shared across them.

Frequently asked questions