Setup guide

Feature flags from your AI agent

Railflag's MCP server lets Claude Code, Cursor, VS Code and Windsurf create flags, change targeting, roll out to a percentage of users and migrate flags from other tools. Production changes still wait for a person.

The Model Context Protocol (MCP) is an open standard that lets AI agents call tools in other systems. Railflag runs a hosted MCP server at https://mcp.railflag.com/mcp, included on every plan.

What your agent can do

AreaToolsFor example
Flagslist_flags, get_flag, create_flag, toggle_flag"Create a flag for the new checkout and turn it on in Staging."
Targeting and rolloutsupdate_targeting, set_rollout, list_environments"Roll new-checkout out to 10% of users in Production."
History and approvalsget_history, get_change_request"Who changed the pricing flag last week, and why?"
Release Watchsuggest_watch_plan, get_watch, list_watches, stop_watch, monitoring_coverage, connect_monitoring"What should we monitor while this rollout runs?"
Migrationconnect_source, scan_source, plan_migration, apply_plan, get_code_guide and more"Import our Unleash flags into Development, then update the SDK calls in this repo."

Step 1: create an access token

In the dashboard, open Settings → Access tokens and create a token. Choose read only if the agent should look but not change anything. The token is shown once and starts with rf_pat_.

The dashboard also shows a ready-made Claude Code command with your token filled in.

Step 2: add the server to your tool

Claude Code

claude mcp add --transport http railflag https://mcp.railflag.com/mcp \
  --header "Authorization: Bearer rf_pat_…"

Cursor

Add this to ~/.cursor/mcp.json, or to .cursor/mcp.json in a project:

{
  "mcpServers": {
    "railflag": {
      "url": "https://mcp.railflag.com/mcp",
      "headers": { "Authorization": "Bearer rf_pat_…" }
    }
  }
}

VS Code (GitHub Copilot agent mode)

Add this to .vscode/mcp.json. VS Code asks for the token once and stores it securely, so it never lands in the repository:

{
  "inputs": [
    { "type": "promptString", "id": "railflag-token", "description": "Railflag access token", "password": true }
  ],
  "servers": {
    "railflag": {
      "type": "http",
      "url": "https://mcp.railflag.com/mcp",
      "headers": { "Authorization": "Bearer ${input:railflag-token}" }
    }
  }
}

Windsurf

Add this to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "railflag": {
      "serverUrl": "https://mcp.railflag.com/mcp",
      "headers": { "Authorization": "Bearer rf_pat_…" }
    }
  }
}

Other MCP clients work the same way: a remote server over Streamable HTTP at https://mcp.railflag.com/mcp, with the header Authorization: Bearer <token>.

Production stays under human review

  • Approvals: in environments that require approval, usually Production, an agent's change becomes a change request. The agent gets a link and gives it to you; nothing changes until a person approves it.
  • Audit log: every change records the agent and the person it acts for, for example "Claude Code via MCP for Maria K.".
  • No lost edits: targeting changes carry the version they were based on, so an agent can't overwrite a change someone else made in the meantime.
  • Read-only tokens for agents that should only answer questions.
  • Read-only migrations: when importing from Unleash or AB Tasty, Railflag only reads from the other tool and never writes to it.

Prompts to try

  • "List the flags that are on in Staging but off in Production."
  • "Create a boolean flag new-search, turn it on for users whose email ends in @ourcompany.com in Production."
  • "Roll new-checkout out to 25% of users, and suggest what to monitor while it runs."
  • "Connect our Unleash instance, scan it, and import everything into Development. Tell me about any rule that can't be translated."
  • "Rewrite the Unleash SDK calls in this repository to OpenFeature with Railflag."

Frequently asked questions

Can an AI agent change Production flags on its own?
No. In environments that require approval, such as Production, a change from an agent becomes a change request with a link. It applies only after a person opens the link and approves it.
Does the Railflag MCP server cost extra?
No. The MCP server is included on every plan, including the free Starter plan.
Which AI tools work with it?
Any MCP client that supports remote servers over Streamable HTTP with a custom Authorization header, including Claude Code, Cursor, VS Code with GitHub Copilot and Windsurf.
Can I give the agent read-only access?
Yes. Create a read-only access token in the dashboard. The agent can then list flags, read targeting and history, but every change is refused.

Give your agent feature flags

Create a free account, make an access token, and paste one command. Approvals, the audit log and the MCP server are included on every plan.

Start for free