1. Pendo Help Center
  2. Integrations
  3. Pendo MCP

Deliver guides through your AI agent (beta)

Last updated:

You can configure your AI agent to search Pendo guides in response to visitor questions and return interactive walkthrough links directly in the conversation. When a visitor asks how to complete a task, the agent can find a relevant guide and send a link the visitor can select to launch the walkthrough in their browser.

This article explains how to configure the Pendo MCP server to support this use case.

Note: This feature is available in closed beta. To request access, fill out the MCP Support Agents interest form. Your Pendo account representative will follow up to turn it on for your subscription.

How it works

When a visitor interacts with your AI agent and asks a task-based question, the agent calls the getInAppGuidance tool on the Pendo MCP server. Pendo evaluates which guides are available for that visitor based on your existing segment rules and returns matching results. The agent can then surface a permalink that the visitor can select to launch the walkthrough directly in their browser.

Use cases

Use this integration to:

  • Deliver interactive, step-by-step walkthroughs in response to support questions, instead of static text answers.
  • Reduce inbound support volume by giving visitors immediate, contextual help.
  • Respect your existing Pendo segment rules when delivering guides.

Requirements

To complete this setup, you need:

  • This feature turned on for your subscription by your Pendo account representative. To request access, fill out the MCP Support Agents interest form.
  • Subscription admin access to turn on the MCP server setting in Pendo and create a service account.
  • The Pendo MCP server turned on for your subscription. To turn it on, go to Settings > Subscription settings > AI Access and turn on Pendo MCP server (read-only tools).
  • A developer or technical resource to configure the MCP client and update your AI agent's system prompt.

Step 1. Create a service account

A service account provides the OAuth 2.0 credentials that your AI agent uses to authenticate with the Pendo MCP server. Service accounts operate with read-only (Viewer) permissions.

Note: If your organization uses multiple Pendo subscriptions, you need to create a separate service account for each one.

  1. Go to Settings > Integrations and select the Service accounts tab. If this tab isn't available, that means this feature hasn't been turned on for your subscription and you must contact your Pendo account representative or our support team to turn it on.
  2. Select + Create service account.
  3. In the Create service account dialog, enter the following:
    • Name. A descriptive label for this service account, for example, "My AI support agent".
    • Applications. Select the applications you want this service account to access.
  4. Select Create.
  5. In the confirmation dialog, copy the Client ID and Client secret values immediately.

Important: The Client ID and Client secret are displayed only once. If you close this dialog without saving them, you must generate a new client secret or recreate the service account.

Step 2. Configure OAuth 2.0 authentication

Your MCP client must obtain an access token before making requests to the Pendo MCP server. Pendo uses the OAuth 2.0 Client Credentials grant type.

Token endpoint

Use the URL that matches where your Pendo subscription is hosted. This is the same host name you use to sign in to Pendo.

  • US: POST https://app.pendo.io/oauth/v1/token
  • US1: POST https://us1.app.pendo.io/oauth/v1/token
  • EU: POST https://app.eu.pendo.io/oauth/v1/token
  • Japan: POST https://app.jpn.pendo.io/oauth/v1/token
  • Australia: POST https://app.au.pendo.io/oauth/v1/token 
Content-Type: application/x-www-form-urlencoded

Request parameters

Parameter Value
grant_type client_credentials
client_id The client ID from Step 1
client_secret The client secret from Step 1
scope read:me

Example request

curl -X POST "https://app.pendo.io/oauth/v1/token" \
  -H "Content-Type: application/x-www-form-urlencoded" \
  -d "grant_type=client_credentials&client_id={clientId}&client_secret={clientSecret}&scope=read:me"

Example response

{
  "access_token": "eyJhbGciOi...",
  "token_type": "bearer",
  "expires_in": 86400
}

Include the access token as a bearer token in all subsequent MCP requests:

Authorization: bearer {access_token}

Note: Tokens expire after the number of seconds indicated by expires_in. Configure your client to request a new token before the current one expires.

Step 3. Configure the MCP endpoint

Point your MCP client at the Pendo MCP server endpoint for your region. Use the URL that matches where your Pendo subscription is hosted. This is the same host name you use to sign in to Pendo.

  • US: https://app.pendo.io/mcp/visitor/shttp
  • US1: https://us1.app.pendo.io/mcp/visitor/shttp
  • EU: https://app.eu.pendo.io/mcp/visitor/shttp
  • Japan: https://app.jpn.pendo.io/mcp/visitor/shttp
  • Australia: https://app.au.pendo.io/mcp/visitor/shttp

This endpoint uses Streamable HTTP transport. Send JSON-RPC requests to this URL and include the Authorization: bearer {access_token} header from Step 2.

Example: Initialize the MCP session

curl -X POST "https://app.pendo.io/mcp/visitor/shttp" \
  -H "Content-Type: application/json" \
  -H "Authorization: bearer {access_token}" \
  -d '{"jsonrpc": "2.0", "id": 1, "method": "initialize", "params": {"protocolVersion": "2025-03-26", "capabilities": {}, "clientInfo": {"name": "your-agent-name", "version": "1.0"}}}'

A successful response returns the server's capabilities and available tools. If the response includes an Mcp-Session-Id header, include it in all subsequent requests.

Note: After receiving the initialize response, send a notifications/initialized notification to complete the handshake before calling any tools.

Step 4. Configure your AI agent

With authentication and the MCP endpoint in place, update your AI agent's system prompt so it calls the getInAppGuidance tool when visitors ask support-related questions.

Important: Your agent must pass the visitor's Pendo Visitor ID as a parameter when calling getInAppGuidance. Pendo uses this to evaluate which guides the visitor is eligible to see based on your segment rules. Your agent has access to this value because it's the same Visitor ID your customers set through the Pendo Web SDK.

Your system prompt should instruct the agent to:

  1. Recognize support intent. When a visitor asks a "how do I..." question, requests help with a workflow, or asks for guidance on a feature, treat it as a signal to call getInAppGuidance.
  2. Pass the Visitor ID. Include the visitor's Pendo Visitor ID in the tool call so Pendo can evaluate guide eligibility.
  3. Return a text response if no guides match. If no guides are returned, the agent should provide a helpful text-based response rather than returning an empty result.

Tip: The specific phrasing of the system prompt depends on your application's domain and user expectations. Focus on making clear to the model when to call getInAppGuidance and how to pass the visitor's question.

Step 5. Surface guide permalinks

When getInAppGuidance returns results, each guide includes a permalink the visitor can select to launch the walkthrough in their browser.

Permalinks come in two forms depending on the guide's configuration:

  • Full URL. When a guide has a launch URL configured, Pendo returns a complete link ready to use directly.
  • Query parameters only. When no launch URL is configured, Pendo returns query parameters without a base URL, for example: ?pendo=RMz6L_ZaNHIlvzSZ4HHwO4wbGLk&step=oZNPUs6q_zOiwyJSzOeWJ7x6Poo. Your agent needs to append these to the URL of the page in your app where the guide appears.

Update your agent's system prompt to instruct it to:

  1. Check each result for a permalink.
  2. Include the permalink in the response as a clickable link.
  3. Format the link using whichever method your chat interface supports:
    • Markdown: [Guide name](permalink_url)
    • HTML: <a href="permalink_url">Guide name</a>

This lets visitors go from asking a question to launching an interactive walkthrough in a single interaction.

Guide types that don't return permalinks

Some guide types are excluded from getInAppGuidance results, including:

  • Guides set to non-public visibility.
  • Resource Center guides.
  • Guides associated with more than one application.
  • Embedded guides.
  • Guides with no steps.
Was this article helpful?
0 out of 0 found this helpful

Articles in this section