What is Hook's MCP?
Customer success teams increasingly work across multiple tools - their CRM, their support platform, their communication stack. Switching between them to get a full picture of an account takes time CSMs don't have.
Hook's MCP (Model Context Protocol) takes the same intelligence layer that already powers Hook, customer health, risk and upsell signals, metrics, notes, and next best actions, and makes it queryable from anywhere. Instead of that intelligence living only inside the Hook UI, the MCP surfaces it into LLM workflows, your own internal tools, and third-party AI clients like Claude, so your whole organisation can act on Hook data, not just the people logged in.
With the the Hook MCP, you can:
Turn Hook data into customer-ready documents: Generate QBR decks, success plans, and escalation summaries directly from live account data, without reformatting anything by hand.
Connect Hook to tools it doesn't natively support: Blend Hook's health scores, metrics, and risk/upsell signals with what's already in your own tools to build things neither could do alone, and add Hook data into artefacts you've already created.
Surface Hook context where your team already works: Let CSMs and leaders pull account information inside the tools they use daily, without opening Hook.
Intelligence that anyone in your organisation can access. Customer intelligence no longer lives only inside Hook, so the whole organisation can act on it, not just the people logged in.
Related Article: Hook MCP: Use Cases for Customer Success Teams
This guide is for setting up the Hook MCP for the first time.
Speak to your Account Manager for more details about accessing the Hook MCP.
MCP Components
The MCP is made up of three parts:
MCP server - Hook's endpoint that exposes your data and workflows as structured, queryable tools
Connection/authentication layer - authenticates the request and scopes it to your Hook instance
Supported clients - the external AI tools and interfaces, e.g. Claude, Cursor, or your own internal builds, that connect to the server to query Hook
How to Connect Hook to your AI Tool
Follow the steps below to connect Hook to your organisation's AI tool.
Step 1: Generate credentials in Hook
Go to your Profile Settings > Access Tokens in Hook.
Find the option to create a token for MCP Clients.
Give the client a descriptive name (e.g. "Claude") and click Create Client.
Copy the Client ID and Client Secret shown on screen.
Important: Copy your Client Secret immediately as it won't be shown again.
Step 2: Configure your external AI tool
Open the settings menu of your AI tool (for example, Claude) and find the Customise or Connectors area.
Click Add Custom Connector.
Paste your Client ID and Client Secret from Phase 1 into the respective fields.
Enter the Remote MCP server URL exactly as follows:
https://mcp-connector.hook.co/mcpClick Connect to complete the setup.
❓Frequently asked questions
Which AI clients or tools does the MCP support?
It's a standard remote MCP server (streamable-HTTP + OAuth), so any MCP client that supports remote servers with OAuth can connect. Validated: Claude and ChatGPT. Other OAuth-capable clients (e.g. Cursor) are expected to work but aren't formally tested.
Is my data secure when using the MCP?
Authentication: Every connection uses OAuth 2.0 (authorization code + PKCE). Each user signs in through Hook's own login at app.hook.co, the same SSO/credentials as the Hook UI, and approves the connection on a consent screen. The AI client never handles a Hook password; it receives only a short-lived access token (a signed JWT, ~1 hour) that it presents on each request. Tokens are cryptographically verified on every call.
Encryption in transit: All traffic is HTTPS/TLS end to end - the client to Hook connection and Hook's internal service calls.
Isolation: Each token is bound to a specific Hook user and customer instance; requests can only ever read that instance's data. Hook's EU and US data regions are fully separate deployments - a connection to one never reaches the other.
Does the MCP send my customer data to a third-party AI?
Hook never pushes your data to an AI provider on its own. Data leaves Hook only in direct response to a query your connected client makes, and only the specific records the tools you invoked return. When you query Hook through, say, Claude, those results are then processed by that client's LLM under that provider's terms (for Claude, Anthropic's), exactly as if you'd pasted the data into it yourself. Nothing is sent in the background, and nothing is shared with any provider you haven't connected.
We could just build this ourselves with an AI tool like Cursor or Claude. Why use Hook's MCP instead?
A self-built integration means building and maintaining your own connectors, auth, and - critically - recreating the customer-intelligence layer Hook has already built. Hook's MCP gives you maintained, structured access to health scores, risk and upsell signals, metrics, notes, workflows, and playbook actions out of the box, with security and permission controls handled for you. You're querying Hook's pre-built context about customer health, so you get to the useful output in minutes instead of spending time on building and maintenance.
Who in my organisation can access the MCP?
The connector is enabled per-customer by Hook. Once it's turned on for your Hook instance, any of your Hook users can connect, each authenticates as themselves, and sees only what they can already see in Hook (see permissions details below). Setup in the AI client is two-tiered: a workspace/Claude admin adds the custom connector once (the URL step), then each team member connects and authenticates individually.
Is the MCP available on the US instance of Hook?
Yes.
What data can the MCP access?
Through the MCP you can access, in structured form: customer intelligence, metrics, health scores, risk & upsell signals, Hook workflows, customer communications, internal customer notes, playbook actions, user data, account hierarchy, and meeting transcripts. This means, effectively, any data from Hook.
Will the MCP respect my existing Hook permissions?
Yes. Access is scoped to the authenticated user's existing Hook permissions, the token carries that user's identity, and every tool enforces the same per-account access control as the Hook UI. A user can only read accounts they already have access to in Hook; connecting through an AI client grants no extra visibility.