Connect Gmail to ChatGPT: Manage Messages, Drafts, and Settings

If you need to connect Gmail to ChatGPT to manage messages, drafts, inbox settings, or automated thread replies, you need a Model Context Protocol (MCP) server. This server acts as the translation layer between ChatGPT's tool calls and Google's REST APIs. You can either build and maintain this infrastructure yourself, or use a managed integration platform like Truto to dynamically generate a secure, authenticated MCP server URL. If your team uses Claude, check out our guide on connecting Gmail to Claude or explore our broader architectural overview on connecting Gmail to AI Agents.

Giving a Large Language Model (LLM) read and write access to an enterprise email ecosystem like Gmail is an engineering challenge. You have to handle OAuth 2.0 token lifecycles, map massive JSON schemas to MCP tool definitions, and deal with Google's notoriously strict message formatting constraints. Every time Google updates an endpoint or deprecates a field, you have to update your server code, redeploy, and test the integration.

This guide breaks down exactly how to use Truto to generate a secure, managed MCP server for Gmail, connect it natively to ChatGPT, and execute complex workflows using natural language.

The Engineering Reality of the Gmail API

A custom MCP server is a self-hosted integration layer. While the open MCP standard provides a predictable way for models to discover tools, the reality of implementing it against Google's APIs is painful. You aren't just integrating standard CRUD operations - you are integrating the Gmail API, which has highly specific design patterns, error formats, and encoding rules.

If you decide to build a custom MCP server for Gmail, you own the entire API lifecycle. Here are the specific integration challenges that break standard assumptions when working with Gmail:

Threading and RFC 2822 Base64URL Encoding

The Gmail API does not accept a simple JSON payload for message bodies like {"to": "client@example.com", "body": "Hello"}. To draft or send an email, the payload must be an RFC 2822 formatted string, encoded in Base64URL format. Furthermore, if you want an LLM to reply to an existing email and keep it in the same thread, simply passing the threadId is not enough. You must extract the Message-ID from the original email and inject it into the In-Reply-To and References headers of the new raw message string. If your MCP server does not expose these constraints clearly to the LLM, the model will hallucinate standard JSON payloads and the API calls will hard fail.

Partial Updates and Label Management

In Gmail, folders do not exist. Everything is a label. Moving a thread to the archive means removing the INBOX label. Marking a message as read means removing the UNREAD label. To perform these actions, the LLM must call specific update methods and pass arrays of addLabelIds and removeLabelIds. Building the schema translation layer so ChatGPT understands exactly which system labels are available (e.g., TRASH, SENT, IMPORTANT) requires meticulous JSON Schema definitions.

History IDs vs. Standard Pagination

When an LLM needs to monitor an inbox for new messages, standard offset pagination fails. The Gmail API uses a highly specific historyId cursor system. To see what changed, you must pass the startHistoryId from your previous state. The API returns incremental delta states (which messages were added, which labels were removed). Handling this cursor logic and explaining it to an LLM via tool descriptions is notoriously difficult.

Strict Rate Limiting and Quota Usage

Gmail enforces strict rate limits based on usage units, not just request counts. For example, sending a message costs 100 units, while reading a message costs 5 units. A multi-step LLM workflow can burn through quotas rapidly. Truto does not retry, throttle, or apply backoff on rate limit errors. When the upstream API returns an HTTP 429 Too Many Requests, Truto passes that error directly to the caller. Truto normalizes the upstream rate limit info into standardized headers (ratelimit-limit, ratelimit-remaining, ratelimit-reset) per the IETF spec. The caller - whether that is ChatGPT or a custom agent orchestrator - is entirely responsible for handling the rejection and applying retry or exponential backoff logic.

How to Generate a Managed Gmail MCP Server

Instead of forcing your engineering team to build a custom translation layer, handle Base64URL encodings, and maintain OAuth token refresh logic, you can use Truto to generate a managed MCP server.

Truto's MCP architecture derives tools dynamically from the integration's underlying REST resources and documentation schemas. When a tool is called, the MCP router flattens the arguments, maps them to the appropriate proxy API endpoint, injects the refreshed OAuth tokens, and executes the request against Google.

You can generate an MCP server via the Truto UI or programmatically via the API.

Method 1: Generating via the Truto UI

For ad-hoc agent testing or internal workflows, the UI is the fastest path:

1. Log into your Truto dashboard and navigate to the Integrated Accounts page.
2. Select your connected Gmail account.
3. Click the MCP Servers tab.
4. Click Create MCP Server.
5. Select your desired configuration (e.g., allow read-only methods, filter by specific tags, or set an expiration date).
6. Copy the generated MCP server URL (e.g., https://api.truto.one/mcp/a1b2c3d4e5f6...).

Method 2: Generating via the Truto API

For production deployments where you are spinning up AI agents for your end-users, you will generate the server programmatically. Make a POST request to /integrated-account/:id/mcp.

const response = await fetch('https://api.truto.one/integrated-account/<GMAIL_ACCOUNT_ID>/mcp', {
  method: 'POST',
  headers: {
    'Authorization': `Bearer ${TRUTO_API_KEY}`,
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    name: "ChatGPT Gmail Assistant",
    config: {
      methods: ["read", "write"], // Allow CRUD operations
      require_api_token_auth: false
    },
    expires_at: "2026-12-31T23:59:59Z"
  })
});
 
const data = await response.json();
console.log(data.url); // The MCP server URL for ChatGPT

The returned URL contains a cryptographically hashed token that securely scopes all operations to that specific connected Gmail account.

How to Connect the MCP Server to ChatGPT

Once you have your Truto MCP URL, you can plug it directly into ChatGPT. There are two primary methods to establish this connection.

Method 1: Via the ChatGPT UI (Custom Connectors)

If you are using ChatGPT Pro, Plus, Business, Enterprise, or Education accounts, you can connect the server natively through the web interface:

1. Open ChatGPT and navigate to Settings -> Apps -> Advanced settings.
2. Enable the Developer mode toggle (MCP support is currently gated behind this flag).
3. Under MCP servers / Custom connectors, click to add a new server.
4. Name: Enter a recognizable label (e.g., "My Gmail Account").
5. Server URL: Paste the Truto MCP URL you generated in the previous step.
6. Save the configuration. ChatGPT will immediately perform the protocol handshake, request the tools/list, and surface the Gmail capabilities to your model.

Method 2: Via Manual Configuration File (SSE Transport)

If you are using a local agent orchestrator, Cursor, or a framework that requires standard MCP JSON configuration, you can use the Server-Sent Events (SSE) transport wrapper.

Create an mcp.json file in your project directory:

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

This configuration instructs your local client to bridge the standard stdio MCP transport over HTTP POST using SSE, connecting it seamlessly to Truto's hosted JSON-RPC endpoint.

Security and Access Control

Giving an LLM unconstrained access to a user's primary inbox is a massive security risk. Truto's MCP servers provide strict boundaries enforced at the infrastructure level. You control these settings when creating the server:

• Method Filtering: Use config.methods to restrict the server to specific operation types. Setting methods: ["read"] ensures the LLM can only execute get and list operations (reading emails), preventing it from accidentally sending emails or deleting threads.
• Tag Filtering: Use config.tags to group tools. If you only want the LLM to manage settings, you can pass tags: ["settings"], blocking access to the raw message payload endpoints.
• Expiration (TTL): The expires_at field sets a strict Unix timestamp for the server. Once the time is reached, an internal durable alarm fires, automatically purging the token from the KV store and terminating the server access permanently. Ideal for temporary agent sessions.
• API Token Authentication: By default, possession of the MCP URL is enough to call tools. For strict enterprise deployments, setting require_api_token_auth: true forces the client to also pass a valid Truto API token in the Authorization header, adding a second layer of identity verification.

Gmail MCP Hero Tools

When ChatGPT connects to your Gmail MCP server, it inherits a dynamic toolset based on the integration's schemas. Here are the highest-leverage tools available for email automation.

list_all_gmail_threads

Retrieves a paginated list of email threads in the user's mailbox. This is the primary entry point for inbox triage.

Usage note: The response only includes the thread IDs and a short snippet. The LLM must pass exact cursor values back to fetch subsequent pages. To read the full contents of a thread, the model must follow up with a call to get_single_gmail_thread_by_id.

"Scan my inbox for the 10 most recent email threads. Identify any threads that look like urgent client requests and give me their thread IDs."

get_single_gmail_thread_by_id

Fetches the complete contents of a specified thread, including all historical messages, labels, headers, and body content.

Usage note: The format parameter dictates how much data is returned (e.g., full, metadata, minimal). The LLM will use this tool to build conversational context before drafting a reply.

"Fetch the full thread for ID '18f8a9b2c4e5d6f7'. Summarize the back-and-forth conversation and tell me what the client is waiting for me to deliver."

create_a_gmail_draft

Creates a new draft message without sending it. This is significantly safer than granting the LLM direct send permissions.

Usage note: The LLM must properly format the payload using RFC 2822 standard and Base64URL encoding. It is highly recommended to have the model create drafts for human review rather than executing live sends in production.

"Based on the thread we just reviewed, create a draft reply to the client. Apologize for the delay, attach the revised project timeline, and leave a placeholder for me to add the final pricing details."

gmail_drafts_send

Sends an existing draft from the user's mailbox.

Usage note: This tool requires the id of the draft you wish to send. It is often used in a multi-step workflow where an agent creates a draft, a human approves it, and the agent then executes the send command.

"I have reviewed the draft with ID 'draft-987654'. The content looks good. Go ahead and send it."

update_a_gmail_thread_by_id

Modifies the labels applied to a specific thread. This is how you implement inbox zero workflows like archiving or marking as read.

Usage note: To archive a thread, instruct the LLM to pass "UNREAD" or "INBOX" in the removeLabelIds array.

"Take thread ID '18f8a9b2c4e5d6f7' and archive it by removing the 'INBOX' label. Then mark it as read."

list_all_gmail_settings_filters

Retrieves all automated message filters and rules configured for the Gmail account.

Usage note: Highly useful for IT admins or agentic workflows auditing inbox rules to ensure critical vendor emails aren't being automatically routed to the trash or spam folders.

"List all the active email filters on my account. Identify any filters that automatically delete incoming messages or forward them to external addresses."

To view the complete inventory of available proxy endpoints, schemas, and required parameters, refer to the Gmail integration page.

Workflows in Action

Exposing these tools to ChatGPT enables complex, multi-step email management routines that previously required heavy custom scripting. Here are two real-world scenarios.

Scenario 1: Automated VIP Triage and Response Drafting

Sales leaders and executives often lose critical communications in the noise of a busy inbox. You can instruct ChatGPT to actively triage incoming messages and queue up context-aware drafts.

"Check my recent emails for any messages from the 'acmecorp.com' domain. If you find a new thread, read the full history, summarize their current pain point, and create a polite draft reply acknowledging receipt and promising a follow-up by tomorrow."

Step-by-step execution:

1. list_all_gmail_threads: The LLM queries the inbox, searching the metadata snippets for the target domain.
2. get_single_gmail_thread_by_id: For the matching thread, the LLM fetches the full message history to understand the client's actual request.
3. create_a_gmail_draft: The LLM synthesizes a response, formats it into the required RFC 2822 structure, encodes it, and posts the draft via the API.

Outcome: The executive opens Gmail to find a ready-to-review draft sitting in the thread, saving them 10 minutes of context gathering and typing.

Scenario 2: Inbox Zero and Automated Archiving

IT teams and developers receive hundreds of automated system alerts, Jira notifications, and Datadog warnings daily. ChatGPT can act as an intelligent filter that cleans up the noise.

"Look at my unread threads. Find any threads that are automated server alerts or newsletter digests older than 3 days. Mark them as read and archive them immediately."

Step-by-step execution:

1. list_all_gmail_threads: The LLM fetches threads with the UNREAD label, filtering by date and sender context.
2. get_single_gmail_thread_by_id: The LLM inspects borderline threads to confirm they are actually automated digests and not direct human requests.
3. update_a_gmail_thread_by_id: For the matching threads, the LLM passes an array to removeLabelIds containing ["UNREAD", "INBOX"], effectively archiving the thread and marking it read.

Outcome: The user's inbox is cleared of low-priority noise, leaving only actionable human correspondence.

The Shift to Managed Infrastructure

Building an AI agent that can reliably parse Base64URL email payloads, handle standard IETF 429 rate limit responses, and accurately manage Gmail labels requires serious engineering overhead. Maintaining the schemas to ensure the LLM understands exactly how to format the data takes time away from your core product.

By leveraging a managed MCP server through Truto, you abstract away the API maintenance, the token refreshes, and the schema generation. You provide ChatGPT with a secure, dynamically generated URL, and your engineering team can focus on designing better agentic workflows instead of parsing Google's REST documentation.