Skip to main content

Connecting Knock2 to AI Tools (MCP)

Use the Knock2 MCP server to bring your visitor intelligence into AI tools like Claude and Cursor — and let your AI assistant take action: reach out on LinkedIn and email, push contacts to your CRM, and add leads to your outbound sequences.

Written by John DiLoreto

Knock2 includes a built-in MCP (Model Context Protocol) server, which lets you securely connect Knock2 to any MCP-compatible AI tool — like Claude, Claude Code, or Cursor. Once connected, your AI assistant can answer questions about your identified accounts, contacts, and visitor activity in plain language — and, with your permission, take action on them too.

This article covers getting connected and how permissions work. For what you can do once connected, see the companion guides: Asking Knock2 Questions in Claude (Reading Your Data) and Taking Action from Claude (Outreach, CRM, Enrichment & Buying Committee).

Reading your data

Through the MCP connection, your AI tool can:

  • Search your identified accounts and contacts — with filters, lead-score ranking, and date ranges that count real visits

  • Pull detailed account and contact records (lead scores with reasoning, UTMs, page visits, CRM deals, and more)

  • Review visitor activity and authoritative summaries over any time period

All data stays scoped to your workspace — the connection only ever exposes your own Knock2 data. Full guide →

Taking action

With write access enabled (see Permissions below), your AI assistant can act on what it finds — the same actions available as buttons in your dashboard:

  • LinkedIn outreach — check whether you're connected with a contact and send a connection request if not, send connection requests with a note, and send LinkedIn messages or InMail.

  • One-off email & LinkedIn touches — draft a personalized message for a contact, then send it. Drafting and sending are separate steps, so there's always a reviewable draft before anything goes out.

  • Add contacts to your outreach tools — list your connected Apollo, Lemlist, Outreach, HeyReach, or Amplemarket sequences and campaigns by name, and enroll contacts into them.

  • Push contacts to your CRM — create or update records in HubSpot, Salesforce (as Leads or Contacts), Unify, or Attio, including firmographics.

  • Discover the buying committee — find new decision-makers and influencers at any account, matching the roles configured in your Settings (2 credits per new contact found; finding nothing costs nothing).

  • Notifications — post a contact card to Slack, or send data to any webhook (for example, a Zapier hook).

  • Data quality — enrich a contact or account with missing details, validate an email address, or refresh CRM deal data for an account.

For example, you can ask Claude: "Find everyone from enterprise SaaS companies who visited our pricing page this week, add anyone not already in HubSpot, and send connection requests to the two most senior people." Full guide →

How to connect

Claude (recommended):

  1. In your Knock2 dashboard, go to Settings → Integrations → MCP and copy the connector URL.

  2. In Claude, go to Settings → Connectors → Add custom connector, name it Knock2, and paste the URL.

  3. Sign in with your Knock2 credentials and approve the requested access. No API key needed.

Claude Code, Cursor, and other MCP clients:

  1. On the same MCP settings screen, click Generate Key to create an MCP API key.

  2. Add the connector URL and key to your tool's MCP configuration (the settings screen shows ready-to-paste snippets for each tool).

You can revoke a key at any time from the same screen — this immediately cuts off access for anything using it.

Permissions, limits, and safety

  • Read-only by default. New connections and API keys can only read data. Write access is granted per category (email sending, LinkedIn, CRM, sequences, notifications, enrichment) — one category never implies another.

  • Admins control write access. Only workspace admins can enable actions — either when approving a Claude connection, or on an API key. Team members connecting their own AI tools get read-only access.

  • Per-tool controls in your AI client. Claude groups Knock2's tools into read-only and write sections in the connector's permission settings, and shows every write action individually — so you can set each one to always allow, ask every time, or deny.

  • Daily limits. Every action type has a sensible daily cap (LinkedIn actions have the strictest), and failed actions don't count against it — so an automation can never run away with your outreach.

  • Everything is attributed and logged. Actions are performed as the user who created the connection, and every action is recorded. Ask your assistant "what are my remaining action limits?" any time.

Already connected to Claude?

Connections made before actions were available will keep working for reading data, but need a quick refresh to enable actions: in Claude, go to Settings → Connectors, remove the Knock2 connector, and add it again. Your Knock2 MCP settings page will show a banner if your connection needs this.

Keeping it secure

Treat your MCP key like a password. Anyone with the key and connector URL can access your Knock2 data — and take any actions the key allows — so only share it with trusted internal tools, and revoke and regenerate if you believe a key has been exposed.

Questions?

Need help getting connected? Message us in Intercom or email support@knock2.ai.

Did this answer your question?