Lead generation

How to connect LinkedIn to ChatGPT to find leads and run outreach (traxy MCP setup)

Connect ChatGPT to traxy through the traxy MCP to find high intent LinkedIn leads and run LinkedIn and email outreach from chat, with your yes first.

By the traxy teamPublished Nov 12, 202616 min read

TL;DR: You don't hand ChatGPT your LinkedIn login. You connect ChatGPT to traxy through the traxy MCP (endpoint https://api.traxy.ai/api/v1/mcp, browser OAuth sign-in), and traxy brings the LinkedIn side: high intent leads found from social intent on LinkedIn, and LinkedIn and email outreach sent from the sender accounts you connect in traxy. traxy's documented setup is for the ChatGPT desktop app and Codex (CLI and IDE extension), where you add a Streamable HTTP server named traxy. ChatGPT on the web uses plugins, and a desktop or CLI setup doesn't carry over to it. Once connected, ChatGPT can search 71M+ scanned LinkedIn likes, posts and comments, read your qualified leads with the post behind each one, and draft, test and launch a campaign, asking for your yes before anything launches or sends.

Checked November 12, 2026. traxy wrote this guide and makes the traxy MCP. traxy's steps come from the traxy docs, linked where they appear. ChatGPT's side comes from OpenAI's own Model Context Protocol guide (checked November 2026). Client menus change between versions, so follow OpenAI's guide when its screens differ from what you see here.

How do I connect LinkedIn to ChatGPT so it can find leads and run outreach?

Connect traxy to ChatGPT as an MCP server, sign in to traxy in the browser, and choose the workspaces ChatGPT may use. ChatGPT then works with traxy's LinkedIn leads and campaigns, and the messages go out from the LinkedIn accounts and mailboxes you connected in traxy.

The short version, with each step explained further down:

Open the guide in traxy. Go to Integrations and select ChatGPT. The connection page that opens is titled Codex and gives app and terminal setup instructions (traxy docs: connect ChatGPT or Codex).
Add the server. In the ChatGPT desktop app, the Codex CLI or the Codex IDE extension, add a Streamable HTTP server named traxy with the URL https://api.traxy.ai/api/v1/mcp. This is the setup traxy documents; ChatGPT on the web uses plugins, and this setup doesn't carry over to it.
Sign in. Complete the browser OAuth sign-in to traxy and choose the default workspace, any other workspaces and the actions you grant.
Check it. Ask ChatGPT to list your authorized traxy workspaces and read a few leads from the one you pick.
Find leads. Describe the people and the need in plain English, and traxy's Social Intent Search returns people with the LinkedIn post or comment behind each one.
Run the outreach. Ask ChatGPT to draft a campaign, test it on sample leads with nothing sent, show you the launch, and launch it only after you say yes.

That is the whole job: traxy runs your entire LinkedIn GTM end to end, from finding high intent leads from social intent to enriching and researching each one, to running relevant, personalized LinkedIn and email outreach to them on autopilot. Autopilot starts only after you approve: you test the campaign on sample leads and approve the launch, and then it runs on its own, or you approve every message in a review queue. The traxy MCP lets you do all of it from ChatGPT instead of the traxy app.

What does the traxy MCP give ChatGPT that it can't do alone?

It gives ChatGPT a source of real LinkedIn leads and a way to send. On its own, ChatGPT can search public web pages and draft messages (OpenAI: searching the web with ChatGPT), but it can't show you who on LinkedIn engaged with the problem you solve, and it doesn't send anything. We cover that gap in more depth in can ChatGPT find leads?

Job	ChatGPT alone	ChatGPT with the traxy MCP
Find people talking about your problem on LinkedIn	No live list; names it suggests need checking row by row	Social Intent Search across 71M+ scanned LinkedIn likes, posts and comments
Check fit	A guess	Only people with a 50%+ ICP match count as leads
Show the reason	No source post	The post or comment behind every lead, labeled Warm, Related or Activity
Contact details	None it can verify	Email or phone lookups on the leads you choose
Research each lead	Public web pages you point it at	A campaign researches each lead for the facts you ask for
Send LinkedIn and email outreach	Drafts only	Launches a traxy campaign from your connected sender accounts, after your yes
Handle replies	No	One inbox across LinkedIn and email; reply from chat after your yes

From ChatGPT you can search social intent, list and research your leads, start a discovery run, qualify the audience of one LinkedIn post, update your ICP and tracked profiles, and send leads to a webhook, Clay, n8n or Zapier destination. You can also run campaigns: draft, test on sample leads, launch, manage sequences and sender accounts, send one-off messages and reply in the inbox. The AI asks for your yes before anything sends or launches. The full list of jobs is in what the traxy MCP server can do.

Your connection only sees the tools your grant, your role and your workspace access allow. The tool list a client gets is filtered to that grant, and each call checks permission again in the workspace it targets (traxy docs: MCP workspaces and protocol).

How do you set up the traxy MCP in ChatGPT or Codex?

Add one Streamable HTTP server named traxy with the URL https://api.traxy.ai/api/v1/mcp, then authenticate in the browser. Where you add it depends on the client: the ChatGPT desktop app, the Codex CLI and the Codex IDE extension share one setup. traxy's documented setup covers these clients; ChatGPT on the web works differently (see below).

What do you need before you start?

A traxy workspace you can sign in to.
The ChatGPT desktop app, the Codex CLI or the Codex IDE extension. OpenAI lists Streamable HTTP servers with OAuth authentication among the MCP features its desktop app, CLI and IDE extension support (OpenAI MCP guide).
No API key. The MCP connection uses OAuth, and a traxy REST API key is not accepted as an MCP credential (traxy docs: connect an AI client). Leave bearer-token and custom authorization-header fields empty.
For outreach later, the LinkedIn accounts and mailboxes you send from, connected in traxy as sender accounts. You can add them now or when you build your first campaign.

How do you add traxy in the ChatGPT desktop app?

OpenAI's guide gives these steps for the ChatGPT desktop app (OpenAI MCP guide, checked November 2026):

Open Settings, then select MCP servers. traxy's docs note that, depending on your client version, this can be the MCPs tab under Plugins.
Select Add server.
Enter the name traxy, choose Streamable HTTP and paste https://api.traxy.ai/api/v1/mcp.
Save the server, then select Restart.
Select Authenticate when the server list shows that traxy needs OAuth sign-in.

OpenAI's guide also says you can type /mcp in the composer to view connected servers.

How do you add traxy in the Codex CLI?

With Codex installed, add the server (traxy docs: connect ChatGPT or Codex):

codex mcp add traxy --url https://api.traxy.ai/api/v1/mcp

Complete browser authorization if it opens. If authentication is still required, run:

codex mcp login traxy

Run codex mcp list to see your configured servers. The Codex CLI is OpenAI's tool; traxy only provides the endpoint it connects to.

How do you add traxy in the Codex IDE extension?

Open the extension's settings, find MCP servers (or the MCPs tab under Plugins), add a Streamable HTTP server named traxy with the same URL, save it, restart the connection if asked, and select Authenticate. OpenAI says the ChatGPT desktop app, the Codex CLI and the IDE extension share MCP configuration for the same Codex host, so once traxy is set up in one, you can switch between them without redoing it.

What about ChatGPT on the web?

traxy's documented setup is for the ChatGPT desktop app and Codex (the CLI and the IDE extension). ChatGPT on the web uses its own plugin connection flow, and a server you configured in the Codex CLI or the desktop app does not carry over to the web client (traxy docs). OpenAI's guide says ChatGPT web uses remote MCP tools through plugins, and that a remote MCP server alone doesn't make a plugin web-compatible (OpenAI MCP guide). If you want traxy in ChatGPT, use the desktop app.

When the browser sign-in opens, you choose the default workspace, any additional workspaces and the actions you want to grant, then approve (traxy docs: authentication and scopes). A few rules worth knowing:

One connection can authorize up to 50 workspaces you select. Calls that don't name a workspace go to the original default.
Grants don't grow on their own. To add a workspace or more actions later, authorize again.
Reading leads and running campaigns are separate permissions. Running discovery, campaigns, outreach and inbox actions sits under the agent:run scope, and contact details stay hidden without contact permission. Grant what you plan to use.

How do you check the connection works?

Ask ChatGPT to list your authorized traxy workspaces, then read a small lead sample from the one you pick. If both come back, the connection, the sign-in and your read permissions are working.

"List my authorized traxy workspaces, then show five qualified leads from the workspace I select."

Look at the result before you ask for anything that changes data or starts discovery (traxy docs: confirm the connection). Then:

Name the workspace. If more than one is connected, say which one in each request. traxy has no server-side "active workspace", so a call that leaves it out goes to the default, not to the one you used last (MCP workspaces).
Check one lead's evidence. Ask: "Show the signals for {lead name} with the source links." You should see the post, comment, like or repost that put the person on your list, with its date.
In Codex, codex mcp list shows whether traxy is configured. In the desktop app, /mcp shows connected servers.

Describe the person and the need in plain English, and traxy searches 71M+ scanned LinkedIn likes, posts and comments and returns people with the post or comment behind each one. You see why each person is there before you decide to contact them (Social Intent Search).

A first search looks like this:

"Search traxy social intent for founders looking for AEO help. Show each person's role, company and the post or comment behind them."

Then narrow it down in the same chat:

Read the reason. Open the post or comment behind the people you like the look of. Check the source and the date before you use it in a message.
Check fit. In your workspace, only people with a 50%+ match to your ICP count as leads, and each lead is labeled Warm, Related or Activity by how strong the evidence is. The match score measures fit, not the chance that someone buys.
Tune the ICP if results are off. Ask ChatGPT to show your active ICP and Signal Brief, propose one change and wait for your OK. traxy keeps the revision history.
Add more sources. Ask ChatGPT to track a competitor's or creator's LinkedIn profile, so traxy keeps checking the people who engage with its posts. Or paste one post URL and ask traxy to qualify its audience. Or start a discovery run for new people who fit your ICP and signals. Discovery and post qualification ask you to confirm before they use capacity.
Enrich who you'll contact. Ask for a work email or phone lookup on the leads you want. Lookups run when you ask, and Auto-Enrich in traxy covers newly qualified leads.
Save the audience. Ask ChatGPT to save the workspace leads you want as a shared lead view. Campaigns use shared views, so this is the list your campaign will run on.

If you already work in Claude, the same steps are in find LinkedIn leads in Claude with traxy.

How do you build, test and launch a LinkedIn and email campaign from ChatGPT?

Ask ChatGPT to draft a traxy campaign for your shared lead view, test it on sample leads with nothing sent, and show you the launch. When you say yes, traxy runs the LinkedIn and email sequence from your own connected sender accounts.

Connect your senders. In traxy, open Outbound > Sender accounts and add the LinkedIn accounts and mailboxes you send from, or ask ChatGPT to open the connect flow. Accounts are grouped by sender name, and a sequence that uses LinkedIn and email needs a sender name with both channels connected (traxy docs: sender accounts). Each account keeps its own limits, active hours and ramp-up, and a campaign does not bypass them.
Draft the campaign. Name the audience, the offer and the facts you want researched. For example: "Draft a traxy campaign for my {shared view}. Research how each company handles outbound today, then prepare a short LinkedIn connection note, a follow-up and an email step." The campaign researches each lead for the facts you ask for, groups leads by signal and gives each group its own personalized message variant built from the evidence. traxy prepares the drafts; you review them.
Test it. Ask: "Test it on sample leads and show me the research and drafts." A test prepares research and previews without sending anything (traxy docs: create a campaign). Check that the research supports each claim, and look at skipped leads as well as good drafts.
Review the launch. Ask for the launch preview. It shows who will join, who is skipped because another unfinished campaign holds them or a completed campaign contacted them before, the sender, the message steps and the delivery mode (traxy docs: review and launch).
Pick the delivery mode. Approve every message in a review queue, or let the approved campaign send on its own.
Say yes. There is no approval card inside ChatGPT, so ChatGPT shows you what it is about to launch and waits for your explicit yes in the chat. Nothing launches or sends before that.

After that yes, the campaign runs on autopilot: it sends on its own because you approved it, or it waits for you on every message if you chose the review queue. Replies, unsubscribes and hard bounces stop the sequence. Approval lets the next step go ahead, but delivery can still wait for sender availability, the schedule or the sender's limits. More on how sending works in LinkedIn and email outreach with traxy.

How do you answer replies from ChatGPT?

Replies from LinkedIn and email land in one traxy inbox, one conversation per lead. Ask ChatGPT: "Show my unread inbox conversations and the thread with {lead name}. I'll reply: {your reply}. Send only after I say yes." You can also pause a lead's sequence from the conversation to take over by hand.

Can you run more than one campaign at once?

Yes. You can run many campaigns at once on different signals, for example one for competitor engagers and one for people posting about the problem you solve. Campaigns run in priority order, so one lead never gets two campaigns at once, and you can pause or change a campaign at any time. For a single message, ChatGPT can send a one-off LinkedIn message, connection request or email to one lead, again after your yes.

For the full workflow from chat across Claude, ChatGPT, Codex, Cursor and Grok, see the LinkedIn GTM MCP page.

What are the common problems and limits?

Most problems are a sign-in that didn't finish, a missing workspace or a missing permission. Each has a short fix (traxy docs: connection and permission issues).

What you see	What to check
ChatGPT or Codex shows no traxy tools	That the OAuth sign-in finished and the connection is enabled. In Codex, run `codex mcp login traxy`. In the desktop app, restart the server and select Authenticate.
It works in Codex but not on chatgpt.com	A desktop or CLI setup doesn't carry over to ChatGPT on the web, which uses plugins. traxy's documented setup is for the ChatGPT desktop app and Codex, so use one of those.
The client asks for a token or API key	Leave bearer-token and custom header fields empty. REST API keys do not authenticate MCP.
A workspace is missing	Authorize again and select it. Grants don't expand when you join another workspace.
A tool is missing	The action wasn't granted, or your role or workspace access doesn't allow it. Reconnect only when you need new consent.
Results come from the wrong workspace	Name the workspace in each request. There is no active-workspace switch.
A message shows "May have been sent"	That is not a failed send. Check the lead's status before trying again, so the person isn't messaged twice.

Limits to keep in mind:

ChatGPT doesn't send on its own. Messages go out through traxy from the LinkedIn accounts and mailboxes you connected as senders, and sending needs those connected sender accounts.
Engagement is a reason, not a decision. A like or comment shows interest in a topic. It does not show budget or timing, so quote the recorded signal in your message instead of guessing at someone's plans.
Review before you approve. Ask ChatGPT to show proposed ICP or setup changes before applying them, and read the test drafts before you launch.
Client screens can change. If a menu has moved, follow OpenAI's MCP guide for the current steps; the traxy endpoint and sign-in stay the same.

Who is this setup for?

B2B companies of any size that sell on LinkedIn: founders doing their own outbound, sales and GTM teams, and agencies running LinkedIn outreach for clients. It suits people who already spend their day in ChatGPT or Codex and want their LinkedIn leads and campaigns there. If you'd rather click through it, the same workflow runs by hand in the traxy app or through trax, the agent inside traxy. You can mix them: find leads in ChatGPT and review the campaign in the app.

Comparing other LinkedIn MCP servers first? Our LinkedIn MCP servers guide sorts them by which ones find leads and which ones send. Once you are connected, the LinkedIn GTM MCP workflow walks through every step from targeting to replies, with a prompt for each.

FAQ

Does ChatGPT log in to my LinkedIn account?

No. The traxy MCP sign-in connects ChatGPT to your traxy workspace, not to LinkedIn. LinkedIn messages go out from the LinkedIn accounts you connect in traxy as sender accounts.

Do I need a traxy API key to connect ChatGPT?

No. The connection uses browser OAuth. A traxy REST API key is not accepted as an MCP credential, so leave token fields empty.

Can I use traxy in ChatGPT on the web?

traxy's documented setup is for the ChatGPT desktop app and Codex (CLI and IDE extension), which add traxy as a Streamable HTTP MCP server and share that setup. ChatGPT on the web uses plugins, and a desktop or CLI setup doesn't carry over to it.

Can ChatGPT send LinkedIn messages and emails through traxy?

Yes, after your yes. ChatGPT can launch a traxy campaign, send a one-off message or reply in the inbox, and the messages go out over LinkedIn and email from your own connected sender accounts. ChatGPT shows you what will launch or send and waits for your explicit yes.

Will a campaign keep sending without me?

Only if you chose that. Once you test the campaign and approve the launch, it runs on its own; or you pick the review queue and approve every message. Replies, unsubscribes and hard bounces stop the sequence.

Can I connect more than one traxy workspace?

Yes. One connection can authorize up to 50 workspaces you select. Name the workspace in each request; otherwise calls go to the default.

Is this the same as trax?

No. trax is the chat inside the traxy app. The traxy MCP brings traxy's tools, including campaigns, senders and the inbox, into another assistant such as ChatGPT, Codex or Claude.

Which other AI clients can use the traxy MCP?

Claude, Codex, Cursor and Grok, as well as ChatGPT. The traxy docs have setup guides per client (connect an AI client).

Back to all articles