Connect Help Scout to ChatGPT: Manage Tickets and Customer Profiles

If you want to automate your support triage, draft context-aware replies, or update user profiles directly from an AI interface, you need to connect Help Scout to ChatGPT. Doing this requires an integration layer capable of translating natural language tool calls into secure REST API executions. You accomplish this using a Model Context Protocol (MCP) server.

If your team uses Claude instead of ChatGPT, check out our guide on connecting Help Scout to Claude. If you are building headless, multi-agent frameworks using LangGraph or AutoGen, read our architectural deep dive on connecting Help Scout to AI Agents.

Granting a Large Language Model (LLM) read and write access to a production Help Scout environment is an engineering challenge. You must map complex nested data structures to flat tool definitions, handle strict pagination, and securely manage OAuth 2.0 token lifecycles. Rather than building and maintaining this infrastructure in-house, modern integration teams use a managed infrastructure layer like Truto to dynamically generate a secure, authenticated MCP server URL.

This guide breaks down the engineering realities of the Help Scout API, shows you exactly how to generate a managed MCP server, connects it natively to ChatGPT, and demonstrates real-world support automation workflows.

The Engineering Reality of the Help Scout API

A custom MCP server is a self-hosted integration layer. While Anthropic's open standard provides a predictable way for models to discover tools, implementing it against vendor-specific APIs is notoriously difficult. If you decide to build a custom MCP server for Help Scout, you are responsible for the entire API lifecycle.

Help Scout's API is heavily normalized and highly structured, which presents three specific integration challenges that break standard CRUD assumptions.

The Mailbox-Centric Architecture

In Help Scout, almost all operational data is siloed by Mailbox. You cannot simply ask the API for "all active conversations." You must query list_all_help_scout_conversations and pass a specific mailboxId. Furthermore, entities like Folders, Workflows, and Custom Fields are strictly bound to their respective mailboxes. If an AI agent hallucinates a global query without first establishing the mailbox context, the API will reject the request. Your MCP server must explicitly guide the LLM to identify the correct mailboxId before attempting conversation retrieval.

The Complexity of Threaded Conversations

A ticket in Help Scout is not a single body of text. It is a Conversation object that contains an array of Threads. These threads represent individual events - customer messages, agent replies, internal notes, and state changes. When an LLM needs to read the history of a ticket, it must extract the conversation ID and then query the list_all_help_scout_conversation_threads endpoint. Furthermore, replying to a ticket requires generating a completely separate entity using create_a_help_scout_conversation_thread, which has strict requirements for the customer or user sub-objects. Managing this multi-step relational logic via JSON schemas is complex.

Embedded Customer Sub-Entities

Customer profiles in Help Scout are not flat records. Emails, phone numbers, chat handles, and social profiles are embedded sub-entities. While you can overwrite some data via a top-level patch, performing precise CRUD operations on a customer's specific email address requires executing sub-resource endpoints like list_all_help_scout_customer_emails followed by update_a_help_scout_customer_email_by_id. If your AI agent sends a flat string to a nested array field, the API will throw a validation error.

Rate Limits and 429 Errors

Help Scout enforces strict API rate limits (typically 400 requests per minute for standard accounts). Truto does not retry, throttle, or apply backoff on rate limit errors. When the Help Scout API returns an HTTP 429, Truto passes that exact error directly back to the caller.

Truto normalizes the upstream rate limit information into standardized headers (ratelimit-limit, ratelimit-remaining, ratelimit-reset) conforming to the IETF specification. The caller - your AI agent or orchestrator framework - is strictly responsible for interpreting these headers and executing its own exponential backoff and retry logic.

Generating the Help Scout MCP Server

Truto abstracts away OAuth lifecycles and dynamic schema generation. It dynamically generates tool definitions by deriving them from the integration's internal configuration and documentation records. If an endpoint is documented in Truto, it becomes an available AI tool. Tools are generated dynamically on every tools/list request; they are never cached, ensuring your LLM always sees the most accurate API surface.

You can generate an MCP server for a connected Help Scout account using either the Truto UI or the API.

Method 1: Via the Truto UI

This is the fastest method for internal tooling and quick prototyping.

1. Navigate to the Integrated Accounts page in your Truto dashboard and select your connected Help Scout instance.
2. Click the MCP Servers tab.
3. Click Create MCP Server.
4. Define your configuration parameters. You can name the server, filter by specific methods (e.g., read-only), and set an optional expiration date.
5. Click Save and copy the generated MCP server URL.

Method 2: Via the Truto API

For production workflows where you need to programmatically provision agentic access for your own end-users, you generate the MCP server via a REST call to Truto.

Endpoint: POST /integrated-account/:id/mcp

{
  "name": "Help Scout Support Triage Agent",
  "config": {
    "methods": ["read", "write", "custom"],
    "tags": ["support", "crm"]
  },
  "expires_at": "2026-12-31T23:59:59Z"
}

Truto validates that the integration has tools available, generates a secure token, stores it in its high-speed edge KV store, and returns a payload containing the url.

{
  "id": "mcp-7a8b9c0d-help-scout",
  "name": "Help Scout Support Triage Agent",
  "config": { "methods": ["read", "write", "custom"] },
  "expires_at": "2026-12-31T23:59:59Z",
  "url": "https://api.truto.one/mcp/xyz123securetoken"
}

This URL is fully self-contained. It encodes the targeted account, the token, and the tool filters.

Connecting the MCP Server to ChatGPT

Once you have your Truto MCP URL, you must register it with your ChatGPT environment to bring custom connectors to ChatGPT. The MCP router handles the full JSON-RPC 2.0 protocol lifecycle (handshake, tools/list, and tools/call) over HTTP POST.

Method A: Via the ChatGPT UI

If you are using ChatGPT Plus, Team, or Enterprise, you can connect the server directly through the graphical interface.

1. Open ChatGPT and navigate to Settings.
2. Click on Apps and open Advanced settings.
3. Toggle on Developer mode (MCP capabilities are currently gated behind this flag).
4. Under the MCP servers / Custom connectors section, click Add new server.
5. Provide a recognizable name (e.g., "Truto Help Scout Server").
6. Paste the Truto MCP URL into the Server URL field and save.

ChatGPT will immediately send an initialize request to the URL, discover the available Help Scout tools, and render them ready for use.

Method B: Via Manual Config File

If you are running local AI frameworks, Cursor, or building a custom agent wrapper that interfaces with OpenAI's API, you can utilize the standard SSE (Server-Sent Events) transport command. Create an mcp_config.json file in your project directory:

{
  "mcpServers": {
    "help_scout_truto": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-sse",
        "--url",
        "https://api.truto.one/mcp/xyz123securetoken"
      ]
    }
  }
}

Your orchestrator will execute this command, tunneling the JSON-RPC calls directly to Truto's edge proxy network.

Help Scout Hero Tools

When you connect the MCP server, Truto dynamically exposes the Help Scout proxy API as tools. The LLM sees a flat input namespace, which Truto's router intelligently maps to query parameters and JSON body properties based on the underlying schema requirements.

Here are the highest-leverage operations for Help Scout AI integration.

list_all_help_scout_conversations

This tool retrieves paginated lists of conversations. Because Help Scout is mailbox-centric, the AI must provide a mailboxId. It supports powerful filtering by status, tag, assigned_to, and modifiedSince. The schema explicitly instructs the LLM on how to pass cursor values back unchanged for pagination.

"Find all 'active' conversations in the support mailbox (ID: 98765) that are tagged with 'urgent' and have not been updated since yesterday."

get_single_help_scout_conversation_by_id

Retrieves the comprehensive metadata of a single conversation, including its state, current assignee, source, and custom fields. Note that this tool returns the conversation envelope, not necessarily the full text of every message in the history.

"Get the details for conversation ID 112233. I need to know the current status, the assigned user ID, and the values of its custom fields."

list_all_help_scout_conversation_threads

To actually read what a customer wrote, the LLM must use this tool to extract the threads associated with a conversation ID. It returns an array of objects detailing the body text, sender info, attachments, and state changes for the entire ticket lifecycle.

"Pull all the message threads for conversation ID 112233 so I can analyze the customer's initial complaint and the last response from our team."

create_a_help_scout_conversation_thread

This writes a new message into an existing ticket. The LLM must supply the conversation ID, the text payload, and specify whether it is an internal note or a customer-facing reply.

"Add an internal note to conversation ID 112233 summarizing the customer's frustration, and flag that engineering is looking into the bug."

get_single_help_scout_customer_by_id

Fetches a complete 360-degree view of a Help Scout customer. It returns primary fields like name and job title, alongside embedded sub-entities like emails, social profiles, and associated properties.

"Retrieve the full customer profile for customer ID 554433. I need to see their primary organization, background notes, and all registered email addresses."

update_a_help_scout_customer_by_id

Executes a patch operation against a customer profile. This is highly useful for AI agents doing profile enrichment - such as updating a job title or adding a social profile link discovered during a triage workflow.

"Update the profile for customer ID 554433. Change their job title to 'Director of IT' and append their newly discovered LinkedIn URL to their social profiles array."

help_scout_workflows_run

Help Scout features a robust internal workflow engine. Instead of forcing the LLM to execute 10 discrete API calls to assign a ticket, add tags, and change status, you can build a manual Workflow in Help Scout and simply have the LLM trigger it using this tool.

"Execute the manual workflow 'Escalate to Tier 3' (Workflow ID: 8899) on conversation IDs 112233 and 112234."

To view the comprehensive inventory of available Help Scout tools, endpoint schemas, and required parameters, visit the Help Scout integration page.

Workflows in Action

Giving an LLM basic tool access is only the first step. The true power of an MCP server emerges when ChatGPT chains multiple operations together to execute autonomous multi-step logic.

Scenario 1: VIP Triage and Summarization

Support queues fill up fast. You want ChatGPT to identify high-value conversations, read the context, summarize the issue, and prepare it for human intervention.

"Check the primary mailbox for any new conversations tagged 'enterprise-tier'. For each one, read the thread history, write a 2-sentence summary of the technical issue as an internal note, and update the ticket status to 'pending'."

Step-by-step execution:

1. list_all_help_scout_mailboxes: The agent first requests the list of mailboxes to locate the ID for the "primary" inbox.
2. list_all_help_scout_conversations: It queries the API using the found mailboxId, filtering for status=active and tag=enterprise-tier.
3. list_all_help_scout_conversation_threads: Iterating through the returned conversations, it extracts the threads for each to read the actual message bodies.
4. create_a_help_scout_conversation_thread: It formulates a concise summary and posts it back to the conversation as an internal note.
5. update_a_help_scout_conversation_by_id: Finally, it patches the conversation record to change the status to pending.

Result: The LLM successfully navigates the mailbox hierarchy, digests complex thread arrays, and writes actionable context directly into the support interface without hallucinating schema structures.

Scenario 2: Customer Enrichment and Escalation

When a customer reports a critical failure, context is everything. You can instruct ChatGPT to augment the ticket data before escalating it to the engineering queue.

"Look up the customer who created conversation ID 998877. If their job title is empty, update it to 'Unknown'. Then trigger the 'Urgent Engineering Review' workflow for this conversation."

Step-by-step execution:

1. get_single_help_scout_conversation_by_id: The agent retrieves the ticket to extract the primary customer's ID from the primaryCustomer object.
2. get_single_help_scout_customer_by_id: Using that ID, it pulls the full profile to evaluate the jobTitle field.
3. update_a_help_scout_customer_by_id: If jobTitle is null, it fires a patch request to update the profile.
4. help_scout_workflows_run: The agent invokes the requested manual workflow ID, pushing the conversation into the engineering pipeline.

Result: By leveraging Help Scout's native workflows alongside direct API writes, the agent updates system records and triggers complex state changes seamlessly.

Security and Access Control

Exposing your Help Scout instance to an autonomous agent carries inherent risk. Truto's MCP implementation provides strict access controls to limit the blast radius of your AI deployments.

• Method Filtering: You can configure the config.methods property on the MCP token to strictly limit operations. Passing ["read"] ensures the LLM can only execute get and list operations, preventing it from ever modifying tickets or deleting profiles.
• Tag Filtering: By leveraging config.tags, you can restrict the server to only expose tools relevant to a specific domain (e.g., exposing only tools tagged with customers while hiding all conversations endpoints).
• Extra Authentication (require_api_token_auth): If enabled, the bearer of the MCP URL must also provide a valid Truto API token in the Authorization header. This guarantees that URL leaks do not result in unauthorized tool access.
• Automatic Expiration (expires_at): You can set a strict TTL for the server. Once the timestamp passes, Truto's Durable Object alarms automatically clean up the database records and KV entries, instantly severing the LLM's connection to Help Scout.

Scaling Help Scout Operations with AI

Connecting Help Scout to ChatGPT transforms a static system of record into an active participant in your support operations. By relying on a managed MCP server, you eliminate the need to write backoff algorithms, maintain OAuth flows, or constantly update JSON schemas when Help Scout modifies an endpoint.

Truto's documentation-driven dynamic tool generation ensures that your AI agents interact with the exact API surface of your integrated accounts in real time. Your engineering team can focus on designing high-leverage agentic prompts and workflows rather than debugging rate limit headers and broken JSON-RPC connections.