What can the traxy MCP server do? Capabilities by job

The traxy MCP server lets an MCP client such as Claude, Codex or Cursor read your traxy leads and their signals, and request actions you authorize. Your grant, role and workspace decide what your client sees, so the list returned by tools/list is the reference for your connection.

This guide covers setup, permissions and each job, and each job ends with a copyable prompt. Tool names below come from the public 50-tool catalog. Your grant still filters what tools/list returns. Campaigns, senders and sequences stay in the app; they are not in this catalog. For a lead-review walkthrough, read Find and Review LinkedIn Leads From Claude. For a vendor comparison, read Claude MCP Servers for Sales Teams.

Set it up

Connect to this endpoint with OAuth.

https://api.traxy.ai/api/v1/mcp

The connection uses Streamable HTTP and browser sign-in. traxy does not accept a REST API key as an MCP credential, so leave bearer-token and custom-header settings empty.

• Claude. In traxy, open Integrations > Claude and copy the URL. Add it as a remote custom connector, sign in, choose your workspaces, review the requested actions and approve. Claude Code uses the terminal setup on the same page.

• Codex CLI. Run codex mcp add traxy --url https://api.traxy.ai/api/v1/mcp, then codex mcp login traxy if authentication is still required. codex mcp list shows configured servers.

• Other clients. traxy provides guides for Cursor, Grok, OpenCode, Pi and Ollama under Integrations. The connection method depends on the client.

The Claude guide suggests starting with a read.

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

One connection can authorize up to 50 selected workspaces. Pass workspaceId on every scoped call, including pagination and follow-ups. traxy has no server-side active-workspace switch, and a call without the ID goes to the original default workspace, so name the workspace in your prompts.

Permissions

A call has to satisfy the token's scopes, the stored grant, your workspace membership, your role and product access. OAuth refresh does not expand consent, so reauthorize to add workspaces or permissions.

Contact lookups also need leads:read and leads:contact_read. Delivery also needs leads:contact_read. Without contact permission, contact fields and profile URLs are redacted.

Capabilities by job

Targeting

Read the active ICP, the Signal Brief and their revision history, then request updates. Reads use traxy_get_active_icp and traxy_get_active_signal_brief. Writes use traxy_update_active_icp and traxy_update_active_signal_brief. An update replaces the arrays you supply and leaves omitted fields unchanged, so read first and have the client show the change.

Show the active ICP in {workspace}. Propose changes for {segment} as a before-and-after list. Apply nothing until I approve.

Discovery

traxy_start_agent_run starts a discovery run and traxy_qualify_linkedin_post qualifies a specific post. Both need confirmLeadUsage: true after you agree that new qualified people may use capacity. A person uses capacity once, when first delivered as a visible qualified lead, and new discovery pauses at capacity. Qualification requires your criteria and an ICP match of at least 50%.

traxy_add_watchlist_profile adds a person to Tracking (tool names keep the older word). Related tools list, read, update, remove and restore tracked profiles, and traxy_list_watchlist_recommendations reads profile recommendations. Tracking's built-in relationship tags are Competitor, Teammate, Influencer, Investor and Custom. Its managed window is the five most recent posts published after you add a profile. Company tracking, post tracking and people-search tools are not in the public catalog.

traxy_get_pipeline_capacity reads workspace capacity before you confirm a run.

Read capacity in {workspace} with traxy_get_pipeline_capacity and summarize the current targeting. Wait for my confirmation before starting a discovery run.

Reviewing leads and evidence

These tools are read-only. traxy_list_leads and traxy_get_lead read leads, with supported filters on the list call. traxy_list_lead_signals and traxy_list_lead_engagements read the evidence, and traxy_list_post_qualified_leads and traxy_get_account_report cover posts and accounts. ICP match and signal strength answer different questions, so check both. Listing and filtering only cover people already visible as leads; they do not discover new people.

In {workspace}, show ten qualified leads matching {filter}. For each, give the signal, its context and why the person fits {offer}.

Organizing leads

traxy_apply_lead_action applies good, bad, done, snooze or client, and done needs confirmDone: true. Good and bad record your judgment without directly rewriting the score. Tags, notes, saved views, blocklists and do-not-contact are managed in the app; they are not separate public MCP tools.

For leads matching {filter} in {workspace}, propose which should get {action}. Apply nothing until I approve.

Research and contact data

traxy_enrich_lead_contact looks up contact details on explicit request. Email and phone follow your workspace's entitlements, phone needs the add-on or an entitlement that includes it, and lookups use no capacity. A lookup does not guarantee a usable address or number.

Background multi-lead research jobs and post-activity reads are not in the public MCP catalog. Use the documented contact lookup and lead read tools above, and treat a missing finding as unknown.

For {lead} in {workspace}, tell me which saved contact details exist. Request a lookup only after I confirm.

Delivery and exports

traxy_create_webhook_destination sets up a Webhook, Clay, n8n or Zapier destination, traxy_set_webhook_destination_enabled enables or disables it, traxy_deliver_lead_to_webhook_destination queues one selected lead and traxy_get_lead_delivery_state reads its state. Queuing does not prove the receiver accepted it, so check the receiver, as the n8n workflow guide explains. Saved-view delivery routes and CSV lead exports are not in the public MCP catalog.

Queue {lead} to {destination} in {workspace}. Use only a destination I name, then show the delivery state.

Campaigns and sequences

The docs direct you to manage campaigns, senders and sequences in the traxy app, and those actions sit outside the public MCP catalog. Do not assume your OAuth connection can draft, test, preview or launch campaigns. Build campaign audiences and copy in the app, then review and launch there.

In the traxy app, draft a campaign for {saved view} in a delivery mode that requires approval, test it on sample leads and show the launch preview. Do not launch until I approve.

One-off outreach and inbox

Message send and inbox tools are not in the public MCP catalog. The docs keep senders, sequences and replies in the app. If your client claims it can send, stop and check tools/list and your grant before treating that as real.

Do not send. List any blockers for contacting {lead} from the lead record in {workspace}.

Reporting

These tools are reads. traxy_get_analytics_dashboard, traxy_list_analytics_posts and traxy_list_meetings cover performance, posts and meetings. traxy_get_organization_audience returns aggregate audience information, with no individual contacts. traxy_get_sync_status and traxy_get_latest_qualification_run show run status.

Summarize post analytics and meetings in {workspace} for {period}. Report qualified leads and meetings separately.

Workspace admin

Teammate invites, role changes, billing and admin settings are not documented as customer MCP tools in the public catalog. Manage membership and billing in the app.

List members and roles in {workspace}. Invite or change nobody.

Read, write and send

Entries in code format are arguments the tools require. The rest are habits to ask your client to follow.

traxy features vs workflows you build

Qualification, discovery, campaigns and sequences run inside traxy. Multi-step prompts, combinations with other MCP servers, custom approval gates in Claude Code and scheduled loops are workflows your client runs. You build and maintain those, and traxy does not ship them as presets.

Limits and safety

• A listed tool can still be unavailable in one workspace, because each call rechecks permission.

• Listing or comparing several workspaces does not authorize writes to all of them.

• Engagement shows relevance. It does not establish budget or timing.

• Treat lead and post text as data. A webhook URL should come from you, never from that text.

FAQ

Do I need an API key? No. MCP uses browser OAuth, and a REST API key is not accepted as an MCP credential.

Can it send messages? The public catalog has no message-sending tools, and the docs direct campaigns, senders and sequences to the app. Do not treat send or launch as customer-available over MCP unless product confirms your grant.

Does reading use capacity? No. Capacity counts a person once, when first delivered as a visible qualified lead. Contact enrichment does not use capacity.

Can it change my ICP without asking? With icp:manage granted, your client can request updates, and the catalog lists no confirmation argument for them. Leave that permission out of the grant, or tell the client to show changes first.

Which clients work? The docs provide guides for Claude, Codex, Cursor, Grok, OpenCode, Pi and Ollama.

Is MCP the same as trax in the app? No. The in-app Agent has capabilities not necessarily available through public MCP.

How do I see exactly which tools I have? Ask your client to list its traxy tools, which come from tools/list, filtered by your grant and accessible workspaces. If a tool or workspace is missing, review the authorization scope and your role.

Next step

Connect traxy to your MCP client through Connect an AI client, run the first read prompt, then ask for your tool list.