👋 Introduction

mcp-harness-fme is a Model Context Protocol server for Harness FME (formerly Split.io). It exposes the FME REST API as MCP tools, so you can inspect and toggle feature flags, manage rule-based segments, and walk your workspaces and environments directly from Claude, Cursor, or any MCP-compatible client.

One API key, zero config friction. Destructive operations are gated behind an explicit confirm: true guard.

✨ Features

• Feature flag management — list, inspect, create, update, kill, restore, archive, and delete flags across any workspace and environment.
• Rule-based segments — full lifecycle management of segments and their per-environment definitions, including a change-request approval flow.
• Rollout statuses — discover rollout status definitions and filter flags by them.
• Multi-workspace, environment-aware — target any workspace and scope operations to specific environments (staging, production, etc.).
• Safety guards — every destructive operation (kill, delete, archive, disable) requires confirm: true, so there are no accidental kills.
• Modern stack — TypeScript 5+, ESM, native Fetch, Zod schemas, stdio transport.

📦 Install

Add the server to your MCP client with the published package — no clone required:

claude mcp add --transport stdio --scope user harness-fme \
  --env MCP_HARNESS_FME_API_KEY=your_api_key \
  -- npx --yes @kud/mcp-harness-fme@latest

Any MCP client works — point its config at npx --yes @kud/mcp-harness-fme@latest and set the MCP_HARNESS_FME_API_KEY environment variable.

Claude Desktop

{
  "mcpServers": {
    "harness-fme": {
      "command": "npx",
      "args": ["--yes", "@kud/mcp-harness-fme@latest"],
      "env": {
        "MCP_HARNESS_FME_API_KEY": "your_api_key"
      }
    }
  }
}

Restart the client after editing its config.

🔐 Authentication

The server needs a single environment variable:

The server exits immediately at startup if the key is missing — no silent failures.

Getting an API key

1. Log in to your Harness account and open the FME module.
2. Go to Admin Settings → API Keys.
3. Click Add API Key and select type Admin (required for management operations).
4. Copy the key and store it somewhere safe.

Important: Admin API keys have full read/write access to your workspace. Treat them like passwords — never commit them to version control.

All requests are sent to the FME REST API at https://api.split.io/internal/api/v2 with a Bearer token.

🛡️ Safety guards

Every destructive operation is gated behind an explicit confirm: true argument. Calling one without it returns an error instead of performing the action:

This makes it safe to let an AI assistant explore your flags — it cannot kill or delete anything without you (or the model) deliberately passing the confirmation.

🧰 Tools at a glance

The server registers 28 tools across four areas:

• Feature flags — 14 tools plus 3 rollout-status / traffic-type helpers.
• Segments — 9 rule-based segment tools plus the classic list_segments.
• Workspaces & environments — the discovery tools you use to find IDs.
• Troubleshooting — when the server or a request misbehaves.

💬 Example conversation

Once configured, you can drive everything in natural language:

You: List all my workspaces
AI:  Shows all FME workspaces with IDs

You: Search for flags containing 'dark-mode' in workspace ws_abc123
AI:  Returns matching flags

You: What are the targeting rules for 'checkout-redesign' in production?
AI:  Shows treatments, targeting rules, and percentage splits

You: Kill the 'new-payments-flow' flag in production
AI:  Requires confirm=true, then disables the flag

You: Restore 'new-payments-flow' in production
AI:  Re-enables the flag and resumes normal traffic routing