AI agent access · MCP

Use MentionLeads from your AI agent

MentionLeads ships a remote MCP server as another secure way to use the same human-first product. Claude Code, Cursor, and any MCP client can pull scored potential customers, reply drafts, traffic, SaaS revenue metrics, retention cohorts, closed-loop customer journeys, goals, funnels, and live visitors — then turn real user complaints into coding tasks your agent executes on the spot. All tools are read-only: a key never posts anywhere, never spends your scans, never touches your account. Configuration examples and the expanded tool reference live in the official MentionLeads MCP GitHub repository.

Same product. Same plans. Human-controlled.

There is no separate agent account, credit system, or agent markup. The account owner controls billing, permissions, and public actions; the agent uses the plan and limits already attached to that account.

Agent guide (.md)

Connect with OAuth

Smithery and other OAuth-compatible MCP clients need only the endpoint https://mentionleads.com/api/mcp. MentionLeads publishes standard protected-resource and authorization-server metadata, uses authorization code with S256 PKCE, and opens a browser so the account owner can sign in, review scopes, and approve. Never append an API key to the endpoint URL.

Agent-led device setup

An agent without browser-based OAuth can start the connection itself without asking for your password, card, or API key. It receives a short-lived pairing link, you open that link to sign in or create an account, choose the normal plan in Stripe if needed, review the requested read scopes, and approve once. The agent then receives a revocable key directly from the one-time exchange.

  1. The agent creates a 15-minute pairing request:
    curl -X POST https://mentionleads.com/api/agent/device-authorization \
  -H "Content-Type: application/json" \
  -d '{"client_name":"My AI agent","scope":"mcp:read analytics:read"}'
  2. It shows you the returned verification_uri_complete. Only continue if the client name and requested access match what you expect.
  3. While you approve in the browser, the agent polls no faster than the returned interval:
    curl -X POST https://mentionleads.com/api/agent/token \
  -H "Content-Type: application/json" \
  -d '{"grant_type":"urn:ietf:params:oauth:grant-type:device_code","device_code":"mld_…"}'
  4. The key is delivered once to the requesting agent and can be revoked anytime from API & MCP.

Manual setup

  1. Create a normal MentionLeads account. Free stays unchanged; API and MCP access starts on Starter and is also included with Pro and Agency. Then mint a scoped token on the API & MCP page in the app. Enable Potential customer intelligence and, for dashboard tools, Growth analytics. Tokens start with mlk_ and are shown once.
  2. Claude Code — one command:
    claude mcp add --transport http mentionleads https://mentionleads.com/api/mcp \
  --header "Authorization: Bearer mlk_YOUR_KEY"
  3. Cursor (or any client using mcp.json):
    {
  "mcpServers": {
    "mentionleads": {
      "url": "https://mentionleads.com/api/mcp",
      "headers": { "Authorization": "Bearer mlk_YOUR_KEY" }
    }
  }
}
  4. Ask your agent: “who are my potential customers today?”, “what changed my MRR this month?”, or “which funnel step loses the most users?”

Tools

ToolParamsWhat it returns
list_projects—Your projects (id, name, product, platforms). Project ids filter the other tools.
get_leadsprojectId?, minIntent?, limit?Latest scored potential customers: real threads with buying intent. Title, URL, subreddit, intent 0–100, quality, and why it was flagged.
get_reply_draftleadIdThe AI-drafted, human-voice reply (and optional DM) for one potential customer. You review, edit, and post it yourself.
get_product_feedbackprojectId?, limit?Build signals mined from your scans: people discussing YOUR product (user_feedback), competitor complaints (competitor_gap), market pains (market_pain).
get_build_promptleadIdTurns one feedback item into a ready-to-run coding task: complaint verbatim, product context, and a definition of done that ends with replying to the thread.
generate_implementation_planideaPro. Idea → market-validated implementation plan as Markdown: demand evidence from real threads, MVP roadmap, go-to-market, and a ready-to-run build prompt. Rate-limited (5/day).
get_account_status—Your plan and how many scans you have left today.
list_analytics_sites—Analytics websites visible to the token. Returns the site ids used by the analytics tools.
get_analytics_overviewsiteId, days?, filters?Traffic and acquisition plus MRR, ARR, LTV, churn, expansion, contraction, cancellations, upgrades, retention cohorts, and the closed-loop customerJourneys array.
get_live_visitorssiteIdThe current five-minute live visitor snapshot.
get_analytics_goalssiteId, days?Conversion goals with completions, visitors, and attributed revenue.
get_analytics_funnelssiteId, days?Funnels with step counts, drop-off, and conversion rates.
get_analytics_segmentssiteIdSaved universal-filter segments for the website.
get_recent_visitorssiteId, days?, limit?Recent sessions with source, location, device, paths, events, and conversion status.

The loop: complaint → fix → reply

The killer workflow for founders who ship fast: get_product_feedbacksurfaces a user saying “love the tool but there’s no CSV export”, get_build_prompt turns it into a scoped coding task (smallest change, tests, no over-building), your agent ships it, and get_reply_draftwrites the human-voice “just shipped it” reply — which you post yourself. Feedback becomes features while the thread is still warm.

Protocol & security

Accounts, plans, and approval

Enter agent codeSame plans · Scoped read-only access · Revoke anytime