How to set up your ICP, signals and LinkedIn tracking from Claude or ChatGPT

TL;DR: Yes. Connect the traxy MCP to Claude or ChatGPT and you can read and change your ICP (including exclusions), your Signal Brief and your LinkedIn Tracking from the chat, and manage your company and people blocklists, with the AI showing each change and applying it after your yes. ICP and Signal Brief changes keep a revision history, so you can see what changed and when. The same connection then runs the second half of traxy: LinkedIn and email campaigns to the high intent leads that setup finds, sent from your own connected sender accounts once you approve.

This guide walks through the setup from chat step by step: connecting, reading your current ICP and Signal Brief, proposing and confirming changes, tracking people, posts and company pages, blocklisting companies and people, and checking the leads that come back. Tool names come from the traxy MCP tool catalog, and the targeting and Tracking rules from the Lead criteria, Intent signals and Tracking docs, checked November 2026.

Can I set up my ICP, signals and LinkedIn tracking from Claude or ChatGPT?

Yes. With the traxy MCP connected, Claude or ChatGPT can read your active ICP and Signal Brief, propose changes for you to confirm, add, pause, remove or restore tracked LinkedIn profiles, track single posts and company pages, and add companies or people to your blocklist.

Your setup decides who traxy brings you. 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. The first step of that is targeting, and it has four parts:

1. The ICP (Lead criteria). Who fits: roles, departments, seniority and locations for people; size, industry, revenue and funding stage for companies; and exclusions for titles, locations or company traits you never want.
2. The Signal Brief. What makes a lead relevant: your business context, buyer pains, triggers, relevant topics, competitors, relevance rules, and keywords and niches.
3. Tracking. The LinkedIn people, single posts and company pages whose engagers traxy checks against your ICP.
4. Blocklists. Companies (by domain) and people (by LinkedIn profile) you want kept out of your lead results, such as existing customers.

In the app these live under Targeting and Settings > Blocklist. From chat you reach the same settings through the MCP, so you can tune targeting in the same conversation where you review leads and build campaigns. A person still only counts as a lead with a 50%+ ICP match, and every lead keeps the post, comment or like that brought them in. For how those parts combine into one decision, see how traxy decides who counts as a lead.

How do you connect Claude or ChatGPT to traxy?

You add the traxy MCP server once, sign in to traxy in the browser, and pick the workspaces the AI may use. No traxy API key is needed.

1. Claude. In traxy, open Integrations > Claude and copy the MCP URL, https://api.traxy.ai/api/v1/mcp. Add it in Claude as a remote custom connector, start the connection and sign in to traxy.
2. ChatGPT and Codex. In traxy, open Integrations > ChatGPT. The connection page is titled Codex and shows the app and terminal steps for the same URL. Follow the ChatGPT and Codex guide for your client.
3. Choose workspaces and actions. Pick the default workspace and any others, review the actions you grant, and approve. Changing targeting needs the ICP permissions (icp:read, icp:manage), Tracking needs watchlist:read and watchlist:manage, and blocklist changes need leads:manage and workspace:read.
4. Start with a read. Ask the AI to list your authorized workspaces and show five qualified leads from the one you pick. That confirms the connection before anything changes.

Cursor, Grok and other clients that support remote MCP with OAuth connect to the same URL; the AI client docs list the options. If you work in more than one workspace, name the workspace in each request so the AI reads and changes the right one.

Reading your current ICP and Signal Brief in chat

Always start by reading what is live. The AI calls traxy_get_active_icp for your Lead criteria and traxy_get_active_signal_brief for your Signal Brief, and can show them as a plain list.

In {workspace}, show my active ICP and Signal Brief as two short lists: who fits, who is excluded, and which topics, pains and competitors make a lead relevant.

Reading first matters for two reasons. Updates use the current revision, so the AI needs it before it can change anything. And an update replaces any list you supply, while fields you leave out stay as they are. If the AI sends a new list of job titles without the old ones, the old ones are gone. Reading first lets it build the full new list from what is there.

When you read the ICP, check three things:

• Does it describe a buyer, not an audience? The Lead criteria docs give the example that "sales leaders at software companies" works better than a long list of unrelated titles.
• Do the exclusions conflict with the inclusions? An excluded title or location that overlaps an included one quietly shrinks your results.
• Is the Signal Brief specific enough to spot in a post? Topics like "struggling to hire a first SDR" can be recognized in public activity. Topics like "growth" can't.

If you have no Signal Brief yet, traxy_generate_signal_brief generates one, which you then read and edit like any other. In the app, entering your website does the same first draft for both the ICP and signals.

Proposing changes and confirming them, with revision history

Ask the AI for a before-and-after list, read it, then say yes. The AI applies the change with traxy_update_active_icp or traxy_update_active_signal_brief, and traxy keeps the earlier version in the revision history.

Propose changes to my ICP in {workspace} so it targets heads of sales and founders at B2B software companies with 10 to 200 employees in the US and UK, and excludes agencies and students. Show the change as before and after. Apply nothing until I say yes.

A good confirmation step looks like this:

1. The AI reads the active ICP and its revision.
2. It proposes the change as a before-and-after list, field by field, including the full new lists.
3. You correct it in plain English ("keep Canada", "drop the revenue filter").
4. You say yes, and it sends the update with the revision it read.
5. It reads the ICP again to show what is now live.

The same flow works for the Signal Brief. For example, add a competitor so engagement with their name counts as relevant, or tighten a topic that keeps bringing in the wrong problem. The Intent signals docs say when to change which: if results point to the wrong problem, change the topics; if the people themselves are wrong, change the Lead criteria.

To see what changed over time, ask for the history. traxy_list_icp_revisions reads ICP revisions and traxy_list_signal_brief_versions reads Signal Brief versions.

List the last revisions of my ICP in {workspace} and tell me what changed in each one.

This is useful when lead quality shifts and you want to know whether someone on the team changed targeting, or when you test a narrower ICP for a week and want to compare. One thing stays in the app: after you save new Lead criteria there, traxy can check your active leads against the new ICP and offer to move the ones that no longer match to Done. Nothing moves until you confirm that.

Tracking people, posts and company pages from chat

Tracking tells traxy whose LinkedIn activity to watch. traxy reads the engagement on those posts and keeps the people who fit your ICP. From chat there is one tool per kind of target:

• People. traxy_track_profiles tracks LinkedIn profiles, and traxy_add_watchlist_profile adds one selected profile. Each profile gets a relationship: competitor, teammate, influencer, investor or custom. (Some tool names use an older name for Tracking.)
• Single posts. traxy_track_posts adds single posts by URL, and traxy_list_tracked_posts lists the posts you added by link.
• Company pages. traxy_track_companies tracks companies. Tracking a company page finds people who engage with that page's posts. It does not list every employee or every buyer at the company.

In {workspace}, track these three LinkedIn profiles as competitors and this one as an influencer: {URLs}. Then track the company page for {domain} and add this post: {post URL}. Show me what you will add before you add it.

To manage what you already track:

• See it. traxy_list_watchlist_profiles lists tracked people or companies with totals, traxy_get_watchlist_profile reads one, and traxy_get_watchlist_summary reads tracking summary metrics.
• Change or pause it. traxy_update_watchlist_profile updates a tracked profile, and you can ask the AI to pause a profile you no longer want monitored.
• Remove or bring back. traxy_remove_watchlist_profile removes a tracked profile and traxy_restore_watchlist_profile restores a removed one.
• Find more to track. traxy_list_watchlist_recommendations reads profile recommendations and traxy_discover_profiles discovers profiles to track.
• Organize it. traxy_list_tracking_tags, traxy_create_tracking_tag and traxy_recolor_tracking_tag manage Tracking tags.

Two rules from the Tracking docs help you read the results. When you add a profile, the first discovery checks its recent posts for new qualified leads, and those leads use your workspace's capacity. After that, monitoring follows the five most recent posts the profile publishes after you added it. For an older post outside that window, add the post itself. Counts across several posts can include the same person more than once, so review the people in Leads.

Not sure who to add? Competitors, your own founders and teammates, creators your buyers follow and the posts that start debates in your niche are the usual starting points. Who to track on LinkedIn for leads goes through each type and what kind of lead it brings.

Profile views are set in the app: Targeting > Signals has a LinkedIn activity card with a Profile views switch, so the people who view your profile can become leads too.

Blocklisting companies and people from chat

A blocklist keeps companies and people out of your lead results. Use it for existing customers, partners, your own team and anyone else you never want as a lead. From chat:

• Companies. traxy_list_company_blocklist lists blocked domains, traxy_preview_company_blocklist previews a change, traxy_add_company_blocklist adds companies and traxy_remove_company_blocklist removes one.
• People. traxy_list_people_blocklist, traxy_preview_people_blocklist, traxy_add_people_blocklist and traxy_remove_people_blocklist do the same for LinkedIn profiles.

Preview adding these customer domains to the company blocklist in {workspace}: {domains}. Show me the normalized domains, then add them after I say yes.

The preview step is worth keeping. The docs advise checking the normalized domain so you don't exclude an unrelated company by mistake.

What a blocklist does and doesn't do, per the Blocklist docs:

• Blocklisted people and people at blocked companies are left out of normal lead results. They are not moved to deleted leads.
• Removing an entry lets otherwise eligible leads appear again and lets future discovery evaluate them. It does not replay past discovery.
• Lead CSV exports don't apply the blocklist, so check an export's audience separately.
• A blocklist is not the same as Do not contact. To stop messages to someone, use Do not contact (traxy_add_do_not_contact from chat).

An ICP exclusion and a blocklist do different jobs. Exclusions rule out a kind of person or company ("no agencies"). The blocklist rules out named companies and people, such as one customer's domain. More on combining them in how to keep customers, competitors and opt-outs out of your prospect list.

Checking the leads that come back

After a targeting change, look at the leads it brings before you trust it. From chat:

• List and filter. traxy_list_leads lists qualified leads with supported filters, traxy_find_leads searches your existing leads, and traxy_get_lead_counts reads counts by tier.
• Read the evidence. traxy_get_lead reads one lead, traxy_list_lead_signals reads its signals and traxy_list_lead_engagements reads the recorded engagements, so you see the post or comment behind each lead.
• Per source. traxy_get_watchlist_profile shows a tracked profile, and traxy_list_post_qualified_leads lists the qualified leads from one post.
• Give feedback. traxy_apply_lead_action marks a lead good or bad, done, snoozed or client.

Show me the 10 newest leads in {workspace} with their ICP match, signal label and the post or comment behind each. Which ones look like a poor fit, and what in my ICP or Signal Brief would you change?

Read two things on each lead, because they answer different questions. The ICP match says whether the person fits; only people at 50% or higher become qualified leads. The signal label (Warm, Related, Activity or ICP match) says how strong the evidence is. A Warm lead with a weak fit and a perfect fit with only Activity both need a second look.

If the leads are wrong, go back one step: wrong people means Lead criteria, wrong topic means the Signal Brief, wrong source means Tracking. Then let the AI propose the fix and confirm it as above.

For new leads now rather than waiting on Tracking, ask the AI to start a discovery run (traxy_start_agent_run) or qualify one post's audience (traxy_qualify_linkedin_post). Both ask you to confirm that new qualified people may use capacity before they run. To go further on finding and researching leads from chat, see how to find and research LinkedIn leads from Claude.

What happens after setup: outreach to the leads from the same chat

Setup is the finding half. The other half is LinkedIn and email outreach to those leads from your own connected sender accounts, and you can run it from the same Claude or ChatGPT chat.

1. Save the audience. Ask the AI to save the leads you want as a lead view (traxy_save_lead_view). Campaigns use shared views.
2. Draft the campaign. traxy_draft_campaign drafts it. A campaign can research each lead for the facts you choose, group leads by signal and prepare a personalized message variant for each group, across LinkedIn and email steps.
3. Test on sample leads. traxy_test_outbound_plan prepares research and previews on a sample. Nothing is sent.
4. Review the launch. traxy_get_outbound_launch_preview previews the launch; check who joins, who is skipped, the senders and the delivery mode.
5. Say yes. traxy_launch_campaign launches only after you confirm in chat.

Once launched, the campaign runs on autopilot because you approved it: steps go out from your connected accounts on their schedule. If you would rather check every message, choose the delivery mode where you approve each one in the review queue. Either way, a reply, an unsubscribe or a hard bounce stops that lead's sequence, and replies land in one inbox you can answer from chat with your yes before each send. The full list of campaign, sequence, sender and inbox tools is in what the traxy MCP server can do.

Because each campaign can start from a different signal (a competitor's engagers, one creator's audience, a tracked company page), you can run several at once, each with its own targeting behind it.

What are the limits of setting up traxy from chat?

• Your permissions apply. A tool showing in the catalog doesn't mean it works in every workspace. Your role, product access and the scopes you granted decide what the AI can do. Reauthorize to add workspaces or permissions.
• Some steps are documented in the app. The website-based first setup, the Check leads step after an ICP change and the Profile views switch are described as app screens, so use the app for those.
• Lists replace, they don't merge. Ask the AI to read first and show the full new list, every time.
• Tracking is engagement, not a full company list. A tracked company page brings people who engage with its posts.
• Fit is not a purchase decision. A 50%+ ICP match and a Warm signal mean someone fits and showed interest. They don't mean the person is ready to buy.
• Other signals come from research. Hiring, funding and job changes come from research on each lead in a campaign, not a live event feed.

FAQ

Can ChatGPT or Claude change my ICP without asking me?

Ask for a before-and-after first and say "apply nothing until I say yes". traxy's Connect Claude docs suggest asking for proposed changes before applying any action, and ICP and Signal Brief changes keep a revision history, so you can see what changed.

Does updating my ICP from chat remove leads I already have?

No. Saving new criteria doesn't move leads by itself. In the app, traxy can check your active leads against the new ICP and offer to move mismatches to Done, and nothing moves until you confirm.

Can I track a competitor's LinkedIn posts from Claude?

Yes. Ask Claude to track the competitor's profile with the competitor relationship (traxy_track_profiles), or their company page (traxy_track_companies). traxy then keeps the engagers who match your ICP, each with the post or comment behind them.

Can I pause a tracked profile from chat?

Yes. Pausing, removing and restoring tracked profiles all work from your AI chat. Ask the AI to show which profile it will change, then confirm.

What is the difference between the blocklist and Do not contact?

The blocklist keeps companies and people out of your lead results. Do not contact stops outreach to a person. Both can be managed from chat.

Does this work with Cursor, Codex or Grok?

Yes. The traxy MCP works from Claude, ChatGPT, Codex, Cursor and Grok, and other clients that support remote MCP with OAuth can connect with the same URL.

Can an agency set up targeting for several clients from one chat?

Yes, if each client's workspace is granted to the connection. Name the workspace in every request so the ICP, Tracking and blocklist changes go to the right client.

Can I launch outreach to the new leads from the same chat?

Yes. Draft a campaign, test it on sample leads with nothing sent, review who joins, then say yes to launch. It sends LinkedIn and email steps from your own connected sender accounts, and replies, unsubscribes and hard bounces stop the sequence.

Related reading

• The LinkedIn GTM MCP and LinkedIn outbound with traxy
• What the traxy MCP server can do, by job
• How traxy decides who counts as a lead
• Who to track on LinkedIn to find leads
• How to find and research LinkedIn leads from Claude
• How to keep customers, competitors and opt-outs out of your prospect list