@finishkit/mcp

MCP server for FinishKit. Production readiness scanner for AI-built apps. Enables AI agents in Claude, Cursor, Windsurf, and VS Code to check if your code is ready to ship.

What AI Agents Can Do

Quick Start

No API key required to get started. The server starts in setup mode and guides you through connecting your account.

Option A: Browser login (recommended)

npx @finishkit/mcp login

Opens your browser. Sign in with GitHub or Google. Your editor picks up the key automatically. No copy-paste, no config editing, no restart.

Option B: Setup command

npx @finishkit/mcp setup

Auto-detects your editor and configures FinishKit. Or target a specific editor:

npx @finishkit/mcp setup --claude-code
npx @finishkit/mcp setup --cursor
npx @finishkit/mcp setup --windsurf
npx @finishkit/mcp setup --codex
npx @finishkit/mcp setup --vscode

Then ask your AI to scan your project. It will show a setup link if you haven't connected yet.

Option C: Manual configuration

Add the following to your editor's MCP config file:

Claude Desktop (~/.claude/claude_desktop_config.json), Cursor (~/.cursor/mcp.json), Windsurf (~/.codeium/windsurf/mcp_config.json), VS Code Copilot (.vscode/mcp.json):

{
  "mcpServers": {
    "finishkit": {
      "command": "npx",
      "args": ["-y", "@finishkit/mcp"],
      "env": {
        "FINISHKIT_API_KEY": "fk_live_your_key_here"
      }
    }
  }
}

Claude Code:

claude mcp add finishkit -- npx -y @finishkit/mcp

Get an API key at finishkit.app/activate.

Works Without API Key

The server always starts, even without an API key configured. This means FinishKit tools always appear in your IDE.

When called without a key, the finishkit_setup tool creates a browser-based activation link. Click the link, sign in, and your editor picks up the key on the next tool call. No restart needed.

Two tools always work without a key:

• finishkit_setup: Creates a setup link and checks connection status.
• create_project: Returns instructions for connecting a repository through the FinishKit dashboard.

Tools Reference

scan_repo (Primary Tool)

Check if your app is ready to ship. Triggers a production readiness scan on a GitHub repository, analyzing security, deployment, stability, tests, and UI completeness. Returns a prioritized finish plan with all findings. Typically takes 2-8 minutes.

Returns: Finding counts by severity and category, human-readable summary, dashboard URL.

get_scan_status

Check progress of a production readiness scan. Returns current phase and progress percentage.

get_findings

Get the production readiness report with prioritized findings blocking launch. Filter by category (blockers, security, deploy, stability, tests, ui) or minimum severity (critical, high, medium, low).

get_patches

Get auto-generated code patches that fix production readiness issues. Each patch includes a unified diff you can apply directly.

list_projects

List all repositories connected to FinishKit for production readiness scanning. No inputs required.

create_project

Get instructions to connect a new GitHub repository to FinishKit for production readiness scanning. Works without an API key.

request_intelligence_pack

Request a production readiness analysis pack tailored to your technology stack. Returns framework-specific rules, security advisories, and analysis prompts for local scanning.

sync_findings

Sync production readiness findings from a local analysis back to the FinishKit dashboard. Creates a run record and inserts findings with deduplication.

finishkit_setup

Set up FinishKit or check connection status. If not connected, creates a browser-based activation link. If connected, shows available tools. Always works, even without an API key. No inputs required.

Resources Reference

For AI Agents

If you are an AI agent with access to FinishKit MCP tools:

Primary workflow: Scan a repository

1. Call scan_repo with the repository owner and name
2. The tool handles the full lifecycle (trigger, poll, wait for completion)
3. Returns finding counts by severity and category
4. Call get_findings for detailed findings list
5. Call get_patches for generated code fixes

Example: User asks "scan myorg/myapp for security issues"

scan_repo({ repo_owner: "myorg", repo_name: "myapp", run_type: "baseline" })

Then:

get_findings({ run_id: "<from scan_repo response>", category: "security" })

Handling common situations

• "FinishKit is not connected": Use finishkit_setup to get a browser activation link
• "Project not found": The repository must be connected at https://finishkit.app/dashboard first
• "Authentication failed": The API key is invalid. Run npx @finishkit/mcp login to re-authenticate
• "Plan limit reached": User needs to upgrade at https://finishkit.app/dashboard/settings

Key facts

• scan_repo typically takes 2-8 minutes. It blocks until complete, no need to poll separately.
• Findings have severity: critical, high, medium, low
• Findings have category: blockers, security, deploy, stability, tests, ui
• Critical and high findings should be fixed before production deployment

Authentication

The simplest way to authenticate is npx @finishkit/mcp login, which opens your browser and stores the key locally at ~/.finishkit/credentials.

The MCP resolves API keys in this order:

1. FINISHKIT_API_KEY environment variable (highest priority)
2. ~/.finishkit/credentials file (written by login or setup --api-key)
3. No key (setup mode with browser activation link)

To get an API key manually:

1. Visit finishkit.app/activate
2. Sign in with GitHub or Google
3. Copy the key (starts with fk_live_)