Triage your LinkedIn sales inbox with ChatGPT-style prompts

The short answer

An AI assistant can read your LinkedIn inbox, sort conversations into “reply today”, “follow up later” and “no action”, and draft the replies for you to approve. Finchreach provides that access through a remote MCP server (MCP is a standard plug that lets AI apps use outside tools) at https://mcp.finchreach.com/mcp.

ChatGPT can connect to Finchreach directly. Finchreach supports OAuth sign-in, so you add it as a custom connector in developer mode, sign in and approve. No API key needed. OpenAI’s Codex and other MCP clients such as Claude Code or Cursor work too, with OAuth or an API key.

Can ChatGPT read my LinkedIn messages?

Yes. ChatGPT’s developer mode lets you add a remote MCP server as a custom connector with OAuth sign-in, which Finchreach supports. Sources: OpenAI developer mode guide and the OpenAI help center, as of October 2026.

You have three options:

• Use the ChatGPT app with a custom connector. Setup is below; it needs a Plus, Pro, Business, Enterprise or Edu plan.
• Use Codex, OpenAI’s coding agent. Its CLI accepts a bearer token from an environment variable, so it connects to Finchreach with a few lines of config. It also works fine for non-coding tasks like this one.
• Use another MCP client: Claude Code, Cursor, VS Code with GitHub Copilot, Gemini CLI and others.

The ChatGPT guide tracks the details.

Why triage a LinkedIn inbox with an assistant?

For anyone selling on LinkedIn, the inbox mixes real buying signals with recruiter pitches, vendor spam, “congrats on the new role” messages and threads that went cold two weeks ago. Reading all of it every morning is slow, and the conversations that matter get buried.

An assistant connected to your inbox can do the first pass for you:

• List recent conversations with linkedin_get_conversations and open the relevant ones with linkedin_get_messages.
• Label each one: hot lead, needs a reply, follow up later, no action.
• Summarize why, in a line, so you can disagree quickly.
• Draft replies for the ones that need them and queue them with linkedin_send_dm into the existing conversation.

Every reply waits in your approval queue with approval mode on, which is the default. You read the draft in context and approve, or reject it and write your own.

What prompts work for LinkedIn inbox triage?

These prompts are written for ChatGPT, Codex or any other MCP client connected to Finchreach. They read the same in any assistant.

Ask your assistant“Read my 20 most recent LinkedIn conversations. Put each one in a table with: name, company if visible, last message date, who spoke last, and a label: HOT (asked about pricing, a demo or a timeline), REPLY (a question waiting for me), LATER, or NONE. Do not send anything.”

Ask your assistant“For the HOT and REPLY conversations, open the full thread and draft a reply under 80 words. Answer their actual question, propose one next step, no pitch deck language. Show me the drafts first.”

Ask your assistant“Queue the drafts I marked ok with linkedin_send_dm into their existing conversations. Then give me the action ids so I can approve them in Finchreach.”

Ask your assistant“Find conversations where I sent the last message more than 10 days ago and the other person had shown interest. Suggest a short, low-pressure follow-up for each, but only queue the ones I confirm.”

How do I set up LinkedIn inbox triage in ChatGPT or Codex?

ChatGPT: In ChatGPT, open Settings → Apps & Connectors → Advanced settings and turn on Developer mode. Then click Create, name it Finchreach, set the MCP server URL to https://mcp.finchreach.com/mcp and Authentication to OAuth. Create it, sign in to Finchreach and approve. In a chat, turn it on from the + menu under Developer mode. Developer mode needs a Plus, Pro, Business, Enterprise or Edu plan.

Codex, with an API key:

1. Create a Finchreach account and connect your LinkedIn account on the Accounts page.
2. Create an API key on the API keys page and export it: export FINCHREACH_API_KEY=lk_....
3. Add the server to ~/.codex/config.toml as shown below.
4. Start Codex and ask a read-only question such as “list my last five LinkedIn conversations” to check the connection.
5. Step-by-step screens are in the Codex guide and the docs; ChatGPT setup is in the ChatGPT guide.

[mcp_servers.finchreach]
url = "https://mcp.finchreach.com/mcp"
bearer_token_env_var = "FINCHREACH_API_KEY"

With bearer_token_env_var, Codex sends Authorization: Bearer <your key> on every request. Source: Codex MCP docs, as of October 2026.

How do I keep inbox automation from feeling automated?

• Never let it reply to a message it has not read in full. Ask it to open the thread, not just the preview.
• Keep replies short and specific to what the person said. A generic reply in a real conversation is worse than a late one.
• Approve in small batches through the day rather than thirty messages at once.
• Do not use it for cold outreach to strangers at volume. Triage is about answering people who already wrote to you.
• Watch for messages that contain instructions aimed at the assistant (“ignore your previous instructions…”). The approval step is where you catch anything odd.

Is it risky to let an assistant answer LinkedIn messages?

Automating a personal LinkedIn account carries risk under LinkedIn’s User Agreement. Finchreach’s approval queue and a human pace lower that risk; they do not remove it.

Reading your inbox is the low-risk part. Sending is where care matters, which is why every message goes through approval before it reaches LinkedIn.

Frequently asked questions

Does this work in the ChatGPT app? Yes, through a custom connector in developer mode with OAuth sign-in. It needs a Plus, Pro, Business, Enterprise or Edu plan.

Can Codex use my ChatGPT conversation history? No. Codex sees what you give it in the session and what it reads through the Finchreach tools.

Can it send attachments in a reply? No, messages are text only.

Can it message people I have never talked to? linkedin_send_dm can start a new conversation, but for triage you want replies in existing threads. Recruiters handling invitations may prefer the Claude invitation review workflow.

Where can I read more? Start with what a LinkedIn MCP server is, then the docs.