MCP connector

Use Crawler inside ChatGPT.

The whole workflow — interview, Knowledge Core, file previews, analytics — runs in the conversation. You only visit this website for checkout.

Connector URL

https://crawler-presence.lovable.app/mcp

Live health: checking… /api/public/mcp-health

1
Open ChatGPT settings and enable Developer Mode
Developer Mode is what allows a custom MCP connector to be added. Without it the connector form is not available.
2
Add the Crawler MCP endpoint
Create a new connector and paste the deployed HTTPS endpoint below as the MCP server URL. Authentication: none — this MVP is a public, no-auth server.
3
Install the connector
Confirm the tool list. You should see start_interview, continue_interview, analyze_source_url, get_knowledge_core, preview_presence, publish_presence, get_pricing, get_analytics, improve_presence, get_checkout_link and get_status.
4
Switch to ChatGPT Work
@ invocation of connectors is currently documented for ChatGPT Work. Use that surface for the steps below.
5
Type @ and select Crawler
Then simply describe what you do, or paste your website link. Crawler infers your entity type and asks one adaptive question at a time.

Tools

start_interviewwrites session state

Free text (+ optional URL) → session_id, first Knowledge Core, first adaptive question.

continue_interviewwrites session state

Answer → merged Knowledge Core and the next tailored question, or interview_complete.

analyze_source_urlread-only

Reads a public HTTPS page and returns candidate facts with provenance and confidence, or an honest unavailable result.

get_knowledge_coreread-only

The full structured Knowledge Core for a session.

preview_presenceread-only

llms.txt, llms-full.txt, about.md and the relevant markdown and JSON previews.

publish_presencewrites session state

Publishes for real once the draft is linked to a subscribed Crawler account; otherwise returns publish_requires_account and a handoff URL.

get_pricingread-only

Plus $5, Pro $20, Business $80 per month with feature differences.

get_analyticsread-only

Clearly labelled demo analytics for 7/30/90 days, plus an AI summary of recurring questions and gaps.

improve_presenceread-only

Turns an insight into the fields to clarify and one targeted question.

get_checkout_linkread-only

External checkout URL; labelled demo mode when Stripe is not configured.

get_statusread-only

Health and debug output: auth mode, model availability, session store, checkout mode.

What works without an account — and what does not

No authentication on the endpoint. Every tool is public. Anyone with the connector URL can call it, so treat a session token like a shareable link.
ChatGPT account identity is not passed to Crawler. The server cannot see who you are and never treats a ChatGPT account as a Crawler user.
Building and previewing are free and need no Crawler account. Interview, Knowledge Core, corrections, file previews, pricing and analytics questions all run in the conversation.
Drafts are durable but anonymous. Sessions are stored in the database for ~30 days under an opaque random token. They are not owned by anyone until you claim them.
Account linking is required for durable ownership, paid subscription management, private analytics, team access and cross-device recovery. That step happens on this website. Sign in with Google at /auth, and the draft you started in ChatGPT is claimed by your account — after that, publish_presence publishes for real while the subscription is active. Crawler also runs an OAuth 2.1 authorization server with a consent screen, so a future authenticated MCP connector can link accounts directly instead of using the website handoff.
Publishing is the paid step. publish_presence hands off to the website with your draft attached; when Stripe keys are absent the same flow runs in clearly labelled DEMO/TEST mode and no charge is made.
Analytics are labelled DEMO. Crawler never has access to private ChatGPT, Claude or Gemini conversations — only Crawler-internal events and observable reads of published files.