---
name: benchmark-setup
description: Connect Benchmark MCP in the current AI client and verify the workspace. Use for Benchmark setup or connection problems.
metadata:
  title: "Connect Benchmark"
  summary: "Connect your AI client and verify your workspace."
  section: "Content"
  authors: "samarth, bob-studio"
  icon: "connect-plus"
  color: "blue"
  order: "1"
  status: "published"
---

# Connect Benchmark

Connect the current client to `https://api.benchmark.app/mcp` with OAuth.
Complete free setup verification before research.

## Reuse the current connection

Determine whether Benchmark tools are already available in this client.
If available, call `whoami` and `list_operations`.
Verify the workspace identity and available operations.
If both calls succeed for the intended workspace, reuse this connection.
Do not add another connection or change another client.

If the workspace is wrong, ask which workspace the user intends to use.
If the client is unknown, ask which client the user uses.
Use that client's supported setup method.
Do not install another client, overwrite configuration, or purchase a subscription.

## Add a missing connection

For a CLI, read its installed help before a configuration change.
Add only the Benchmark server entry.
Preserve other server entries and settings.
For a browser client, guide the user through its connection settings.
The user completes OAuth sign-in and chooses the intended Benchmark workspace.
Do not request passwords, API secrets, or copied OAuth tokens.

### Claude Code

Run this command in the intended environment:

```sh
claude mcp add --transport http benchmark https://api.benchmark.app/mcp
```

Use `/mcp` in Claude Code to authenticate.
See [Claude Code MCP documentation](https://code.claude.com/docs/en/mcp) if the installed command differs.

### Codex

Use the MCP controls in the current Codex client, or its supported CLI commands:

```sh
codex mcp add benchmark --url https://api.benchmark.app/mcp
codex mcp login benchmark
```

Run these commands in the environment that supplies this client's MCP connections.
A CLI configuration does not necessarily configure a separate web client.

### Cursor

Open Cursor's MCP settings and add an HTTP server with this entry:

```json
{"mcpServers":{"benchmark":{"url":"https://api.benchmark.app/mcp"}}}
```

Merge the entry with existing configuration.
Use Cursor's Connect or sign-in control to complete OAuth.
See [Cursor MCP documentation](https://cursor.com/docs/context/mcp) for current settings.

### Claude, ChatGPT, or another web client

Use the current client's custom connector or MCP app settings.
Set the server URL to `https://api.benchmark.app/mcp`.
Select OAuth when the client requests an authentication method.
Complete Benchmark sign-in, enable the connection, and return to this conversation.
If an administrator must enable custom connectors, explain that requirement and stop until access exists.
If custom MCP is unavailable, explain the limitation without changing clients automatically.
Manual client guidance is at [benchmark.social/mcp#connect](https://benchmark.social/mcp#connect).

## Verify and return to the task

Call `whoami` and `list_operations` through this client's connection.
Report the workspace identity and whether mutations are available.
A read-only connection can still complete setup verification.
Do not treat missing mutation access as permission to change scopes or billing.
If tools remain unavailable, report the failed step and a specific next action.
Do not claim success from configuration text alone.

For a setup-only request, stop after these free verification calls.
For research already requested, return to that task with its supplied sources, context, and approvals.
Use [content-research](https://benchmark.social/skills/content-research/SKILL.md) for a post breakdown or business strategy.
Ask only for missing scope or credit approval before paid work.
Do not require the user to repeat the task or brief.
Do not call `query` or `mutate` merely to test the connection.