promptleads-mcp

The official Promptleads MCP server. Pull a managed client's AI-visibility report and approved answer-first page fixes straight into Claude Code, Cursor, Codex, or Claude Desktop.

Dependency-free, spec-compliant JSON-RPC over stdio. Scope is the client's white-label share token — no database credentials or API keys live in this process, and only human-approved fixes are ever returned.

npx promptleads-mcp

Tools

• get_page_fixes({ shareToken }) — for a code-built site (Claude Code / Cursor / Codex / v0 / Astro / Next.js / Hugo — anything in a Git repo, no CMS): the client's APPROVED answer-first page fixes, ready to apply in the repo. Each returns a title, slug, answer-first sections, the target buyer prompt, and schema.org JSON-LD. Your coding agent writes each page into the repo in the site's own format and opens a PR.
• get_client_report({ shareToken }) — the full report JSON: AI presence with 95% confidence intervals, Maps Share-of-Local-Voice, citation audit, reviews, GBP drafts, AI-readability.
• summarize_client_report({ shareToken }) — a concise human-readable summary.
• get_competitors({ shareToken }) — who the AI engines name instead of the client (share-of-voice leaderboard).
• get_fix_targets({ shareToken }) — missing directory citations, off-site mention targets, AI-readability flags.
• get_attribution({ shareToken }) — AI-referral visitors, conversions, and revenue from the client's own GA4 + Search Console.

The shareToken is the token from a client's white-label report link (/report/<token>), found in Promptleads under Settings → API & MCP or on the client's report page.

Operator tools (with a workspace API key)

Set a workspace API key (pk_live_…, from Settings → API & MCP) as PROMPTLEADS_API_KEY and four workspace-scoped tools unlock — run your whole workspace from your AI tool:

• list_clients() — your managed clients (id, name, city, vertical, report link).
• get_client_visibility({ clientId }) — a client's AI-visibility summary with 95% CIs.
• stage_fix({ clientId, topic?, area? }) — draft an answer-first page fix and stage it for approval.
• list_pending_approvals() — the staged actions awaiting a human decision.

The seatbelt: stage_fix only stages a draft — it never approves, publishes, or sends. A human approves in Promptleads (then it can go live, or be pulled into a repo via get_page_fixes).

Connect

Claude Code

claude mcp add promptleads -- npx -y promptleads-mcp

Cursor / Codex / Claude Desktop

Add to your MCP config (claude_desktop_config.json, ~/.cursor/mcp.json, or the equivalent):

{
  "mcpServers": {
    "promptleads": {
      "command": "npx",
      "args": ["-y", "promptleads-mcp"]
    }
  }
}

Then ask your agent: "Using the promptleads MCP, get the page fixes for share token <token> and apply them to this repo."

To unlock the operator tools, add your workspace API key to env:

{
  "mcpServers": {
    "promptleads": {
      "command": "npx",
      "args": ["-y", "promptleads-mcp"],
      "env": { "PROMPTLEADS_API_KEY": "pk_live_…" }
    }
  }
}

PROMPTLEADS_API_URL defaults to https://promptleads.vercel.app; set it in env only if you self-host.

How it works

It speaks newline-delimited JSON-RPC on stdio (initialize, tools/list, tools/call, ping), thin over the public Promptleads report API. The share token is the only scope — treat it like a link-secret.