Give ChatGPT and Claude the Conversation They Missed

An AI can write a convincing follow-up while missing the one fact that should change it.

Imagine a customer tells you, as you leave a meeting, that their security team must review the product before procurement can proceed. Later, you ask ChatGPT to draft the follow-up from an earlier email thread. The draft asks when they can sign. It sounds professional, but it skips the request the buyer just made.

If you saved that conversation in Draki, you can give the assistant a way to retrieve it. The Draki app works by itself, using OpenAI cloud transcription after your permission. The public app does not automatically recognize speakers. Its optional bracelet keeps capture conveniently within reach, with recording you can turn on or off whenever you want.

Today, Draki connects to ChatGPT, Claude and other compatible AI clients through custom MCP setup. You add the connection in your assistant's settings, sign in to Draki and authorize read access to your synced conversation text and context. Connecting is optional. You can then ask what the customer requested and draft from the answer.

Connect your account

There isn't a Draki plugin in ChatGPT's public directory yet. We're working on one. For now, you connect through custom MCP setup. You don't need to write code: add the address, sign in and approve access.

ChatGPT, Claude and other compatible clients use the same server address:

https://api.draki.ai/api/mcp

1. In Draki, open Profile → Connect with your AI, choose your assistant and copy the server address.
2. Open your assistant's connection settings and add Draki as a custom remote MCP connection using that address. Choose OAuth for authentication. The Draki connection guide has separate setup instructions for ChatGPT, Claude and other clients.
3. Sign in with your Draki email and approve read access to your synced account context.
4. Return to your assistant, enable Draki in the conversation and ask it to use your saved context.

For ChatGPT, start on the web and enable Developer mode if it is available for your account. The OpenAI setup instructions explain where to add a custom MCP connection.

You can also ask ChatGPT to walk you through it:

Help me connect Draki to ChatGPT through custom MCP. Use the guide at https://api.draki.ai/connect and walk me through the setup for my account.

The MCP server and authorization flow are shared; the setup screens are not identical. Availability depends on the client, your plan and workspace settings. Other clients need support for remote MCP over Streamable HTTP with OAuth; support for local MCP servers alone is not enough.

Model Context Protocol, or MCP, is a standard for connecting AI applications to external tools and data. It lets you use Draki's saved context with an AI that supports the protocol, rather than being tied to one assistant.

A pasted server address alone does not grant access. You complete the Draki sign-in and approval for each assistant you connect. Approval grants ongoing read access to your synced account context until revoked or expired; it is not permission for just one question.

Start with the question behind the email

For the customer in our example, ask the assistant to recover the request before it writes anything:

Use Draki to find my conversations with Alex Morgan from September 1 through September 20, 2026. If more than one person matches, ask me which Alex. What did Alex ask us to do before procurement could proceed? Include the conversation and date, and the relevant part of the transcript.

This is easier to check than a broad request to analyze your whole relationship with Alex. You are looking for the security review and anything you promised to send for it.

Open the source if the answer changes what you plan to say. "Security needs to review this" and "security has approved this" lead to different emails. A source reference helps you catch the difference before a fluent draft hides it.

Once the answer is right, continue:

Draft a short follow-up acknowledging the security review and asking what documentation the team needs. Use only the commitments we made in that conversation. Leave the draft here for me to review.

Draki provides the recorded material; your chosen assistant writes the draft. The sales meeting notes article covers keeping requests and commitments clear across several visits.

What the connection can read

The current connection can search synced conversation text by topic and time, retrieve original transcript passages and look up available history for a confirmed person. You can ask about one meeting or follow a topic across several discussions.

New public Draki recordings use OpenAI cloud transcription after your permission. The public app does not automatically separate or recognize speakers. Raw audio, voice samples and voice profiles are not shared through MCP. Conversation text and related context sync to your Draki account, where cloud services support features such as summaries and search. An authorized assistant retrieves results from that synced context. If a conversation is missing, check that its text has synced and that the connection is authorized. A recording can exist on your phone without its context being available to the assistant.

The Draki connection is read-only. Sending emails, updating a CRM and deleting conversations are outside its current scope. An assistant with other connected tools may have separate abilities; authorizing Draki does not grant those permissions.

Use a different assistant without rebuilding the history

You might prefer Claude for reviewing a long discussion and ChatGPT for drafting from it. Authorize each separately and each can retrieve context synced to your Draki account. Their chat histories and built-in memory settings remain separate. These assistants process shared text in their own services; connecting them does not make their models run on your phone.

You can also disconnect either one. That stops future access through the connection; it cannot recall information the provider already received. Read Draki's privacy policy alongside the assistant provider's policies when deciding what to share.

The practical gain is that you no longer have to type out the missing conversation from memory before your AI can help. If you are starting with capture, the conversation app article walks through saving a discussion and using it before the next meeting.

Join the Draki waitlist for product and availability updates.