Fareground MCP

Give your agent the power to simulate. Connect any MCP client to Fareground and your agent can run ready environments, build custom scenarios, run them hundreds of times and read the results — the same work you do in the app, from the agent you already use.

Server URL: https://app.fareground.com/mcp

The server speaks streamable HTTP and signs in with OAuth. The connection acts as your account, with the access you consent to: read to look, read/write to build and run.

Connect

Claude Code

claude mcp add --transport http --scope user fareground https://app.fareground.com/mcp

Run /mcp in Claude Code to sign in. Leave out --scope user to install it for one project only.

Claude desktop and web

Add a custom connector with the server URL, where your plan supports custom connectors, and sign in when asked.

Codex

codex mcp add fareground --url https://app.fareground.com/mcp
codex mcp login fareground

Cursor

Add this to ~/.cursor/mcp.json (every project) or .cursor/mcp.json (one project), then enable it and sign in:

{"mcpServers":{"fareground":{"url":"https://app.fareground.com/mcp"}}}

Any other MCP client

Point it at the server URL as a remote (streamable HTTP) server and complete the OAuth sign-in it opens.

Check the connection

Ask your agent to call whoami. It returns the account and its scopes — check for write access before asking for anything that builds or runs.

What your agent gets

Three things arrive with the connection, not just tools:

• 66 tools in six groups — every one listed with its signature and description in the tool reference.
• The Fareground skill — operating instructions delivered the moment it connects: how to pick an environment, build and check one, run it and report honestly. See skill and guide.
• A guide tool and fareground://skill/... resources — the same documentation on demand, so clients that skip server instructions still get it.

What to ask

Your agent picks the tools; you describe the question. A few that map straight onto Fareground:

• "What happens to churn if we raise prices 8% next quarter?" — a Market session with the price change as a setting. See run a ready environment.
• "Our supplier says we breached the contract. Who's right?" — a Dispute session; your agent writes your side of the case before it starts.
• "Will the Fed cut, hold or hike in December?" — a Council session that debates the question and votes.
• "If our main supplier is two weeks late, do our stores run out?" — a custom scenario built from a supply-chain engine and verified before it runs. See answer a question in one call.

Access, costs and safety

• You decide when money is spent. Your agent prepares and checks a run, shows you what it is set to — length, population, who plays them, key inputs, number of runs — and starts it when you say so. Preparing a ready-environment session is free; starting it is billed.
• Check before a big run. get_balance reads your wallet; get_cost_estimates gives historical estimates with their sample sizes. Estimates are not price quotes or spending caps.
• No double charges on retries. Every paid start takes a UUID request_id. After an uncertain response your agent retries with the same ID and arguments; a new ID means new, separately billed work.
• Read-only means read-only. A read-only connection fails every call that changes or starts anything.
• Long work is durable. A study returns an operation_id at once; your agent follows it with get_custom_study or cancels with stop_custom_study. Wait tools return within 30 seconds, and a timeout never cancels the work.
• Hints for your client. Every tool carries read-only, destructive and retry-safe hints, shown as flags in the tool reference. Authorization is still enforced by the API.

In this section

• Workflows — the step-by-step paths, from a one-call study to a hand-built environment.
• Tool reference — all 66 tools.
• Skill and guide — what your agent reads, and every guide topic.