Any Wingmen account can connect

Use Wingmen from Claude

Connect Wingmen to Claude and it can work with your workspace directly: ask about your pipeline, approve leads, set up and switch on agents, and answer replies, in plain language instead of clicking through the dashboard. It runs over MCP, which is the standard way Claude connects to an outside tool.

It works in Claude on the web, in Claude Desktop and in Claude Code, and the tool set will change as people tell us what is missing.

What Claude can do

Thirty-five tools. Every one is read, write or send, and a connection only gets the kinds you allowed. Nothing reaches another person without your yes: a message goes out only after you have seen its exact text and confirmed it, and an agent starts only after you confirm switching it on. Each confirm carries a signature from the preview you saw, so a changed message or a changed agent is refused, not sent.

Read: twenty-one, and every connection gets them

  • Pipeline status get_pipeline_status

    Where the workspace stands right now: leads by status, how many came in today, the daily lead target, whether discovery and outreach are running, and which list approved leads are saved to. Each LinkedIn account is listed with its last 7 days, its last Lead Radar run, its LinkedIn searches left today, and each source on or off with its last 30 days: found, surfaced, approved and accepted.

  • Leads waiting on you list_pending_leads

    The queue in Lead Radar, highest score first, with enough on each person to make the call. It also reads Lower confidence with the reason each lead is there, the approved leads, and Show removed, and narrows by source or score.

  • Contact search search_contacts

    Find a saved contact or lead by name, company or email. It covers both your saved contacts and the leads the copilot found, and someone who is in both is returned once, tagged with where they came from. You can narrow it to a discovery status (approved, pending, low_confidence, removed), which searches the copilot's leads rather than your saved contacts.

  • Campaign performance get_campaign_performance

    How the agents are doing: leads enrolled, invites sent, acceptances, replies, send failures, and the all-time funnel.

  • Where each agent is agent_progress

    Each agent's next step and the earliest time it can really go out, with the sending hours, limits and pacing applied, and what is holding it back right now. Name one agent to also get its setup and its message steps, or one account to see only its agents.

  • Conversations get_conversations

    Your Wingmen inbox. Filter by sentiment or search by lead name. It also lists the drafts an agent is holding for you, with the reason each is held.

  • One conversation in full get_conversation

    Every message stored for one thread, oldest first, with the lead's card. It does not mark the thread as read.

  • One lead get_lead

    One person's Lead Radar card, the lists they are on, and the stage they are at in each agent: how many steps are done, the last one and the next.

  • Your lists list_lists

    Your contact lists with how many people are on each and which agents work them, and the people on any one list.

  • Activity and Radar notes get_activity

    The activity feed from your dashboard, and what Lead Radar has learned and changed for one account.

  • Automations list_automations

    Each automation, whether it is on, how often it ran, when it last ran and why it last skipped.

  • Your accounts get_accounts

    Each connected account as the Accounts page shows it: status, send window, what it has used of each daily limit, and its LinkedIn searches left today.

  • Workspace settings get_workspace_settings

    Your settings: company details, whether outreach is paused or in dry run, auto-approve, the email channel, notifications, the plan and the members.

  • Usage against your plan check_usage

    Contacts generated, messages sent, AI credits, connected accounts, and how much of each is left, including this month's lead allowance.

  • Your targeting view_icp

    What Lead Radar runs on for one LinkedIn account, field by field, exactly as it is saved.

  • Your LinkedIn connections list_my_connections

    Search the people you are already connected to, by name and headline. It answers from a copy we hold, so it is instant and costs nothing. Every answer says how much of your network has been read so far, so a count is never mistaken for the whole thing.

  • Read in more of your connections sync_my_connections

    Fetches the next few hundred of your own connections into that copy. A large network takes several goes, and it picks up where it stopped. It sends nothing and contacts nobody; it uses your account's own daily allowance for reading your connection list.

  • Who engaged with a post get_post_engagement

    The people who commented on or reacted to a LinkedIn post, yours or anyone's. It contacts nobody, and it spends your account's daily comment-reading allowance, the same one Lead Radar draws on.

  • Find leads find_leads

    Describe who you want. If targeting is set, it tells you what Lead Radar is working on and returns the leads waiting for you. If not, it proposes targeting for you to confirm; nothing is saved and no search runs until you do.

  • Build targeting from a source generate_targeting

    Proposes Lead Radar targeting from your website, a description, a customer list or the text of your deck, next to what is set now. Nothing is saved. Building it uses one AI analysis and a few of the account's LinkedIn searches.

  • Propose a targeting change propose_targeting_change

    Turns a plain instruction such as "stop targeting the US" into a before and after for each field. Nothing is saved.

Write: twelve, only if you allow changes

These change your workspace. None of them sends a message. A read-only connection cannot see or call them.

  • Hand a lead to another account assign_lead

    Gives one person to another of your LinkedIn accounts: that account owns them from then on, and the old owner's agent stops writing to them at once. It sends nothing.

  • Approve a lead approve_lead

    Saves a lead from the queue to your target list. Nothing goes out at that moment. The campaign engine drains that list on its own schedule, under the same pacing, billing and safety limits as an approval you click yourself.

  • Approve a batch approve_leads

    Shows you the exact leads it would approve and where each lands, then approves them only after you say yes, with the signature from that preview.

  • Dismiss a lead dismiss_lead

    Takes a lead out of the queue so it stops coming back. Nothing is deleted, the lead stays on record as removed.

  • Save connections to a list save_connections_to_list

    Puts people from your own 1st-degree connections into a new contact list. A list is only worked once you point an agent at it.

  • Save a post's engagers to a list save_post_engagers_to_list

    Puts people from a post's engagement into a new list, keeping the post and their own comment so outreach can mention it honestly.

  • Apply targeting confirm_targeting

    Applies a targeting proposal you were shown, checked against its signature, and starts that account's Lead Radar run. A run spends real search budget.

  • Create an agent create_agent

    The same save as the New agent wizard. It is created paused, so it sends nothing until you switch it on. Writing its sequence uses AI credits, as it does in the app.

  • Change an agent update_agent

    Shows each field before and after, and saves only after you say yes, with the signature from that preview.

  • Switch an agent on activate_agent

    The step that starts real outreach. It shows you the agent, the account, its send window and limits and anything that would stop it, and starts only after you say yes, with the signature from that preview.

  • Pause an agent pause_agent

    Stops it at once, in one call, because stopping is always safe. Replies still reach Unibox.

  • Switch a Lead Radar source set_source

    Turns one source on or off for one account, the same switch as in Change targeting. It messages nobody.

Send: two, only if you allow sending

The only tools that put a message in front of another person, one person at a time, through the same copy check and daily limits your agents have. Sending is never implied by allowing changes: a connection made before sending existed has to reconnect to get these.

  • Reply in a thread reply_to_thread

    Replies to one person in a thread already in Unibox, from the account the thread arrived in. It shows you the exact message first and sends only after you say yes, with the signature from that preview.

  • Message a connection send_message

    One LinkedIn message to one person: into the thread you already have, or a new chat with a 1st-degree connection. Never a connection request, never email. It shows you the exact message first and sends only after you say yes.

What it cannot do

Claude cannot send a connection request or an email, post or comment, or message more than one person at a time. Those are not exposed to it at all, on any scope, so there is no prompt that reaches them.

Setting it up

1. Claude on the web, the short way

No key to copy and nothing to paste into a config file. In Claude, open Settings, then Connectors, then Add custom connector, and give it this address:

https://app.flywingmen.com/api/mcp

Paste it exactly as it is, starting with https://. Claude will not accept the bare domain. You will land on a Wingmen page that names your workspace and lists what the connection can and cannot do; press Allow and you are connected. If you are not signed in to Wingmen yet it will ask you to, then bring you straight back to that page.

2. Claude Code and Claude Desktop

These use a key instead. In Wingmen, open Settings and pick the Claude / MCP access tab. Give the key a name so you can tell your keys apart later, choose what Claude may do (read only, read and make changes, or that plus sending messages), then press Create key. The key is shown once, so copy it before you close that panel. The same panel hands you both snippets below with your key already filled in.

2a. Claude Code

Run this in your terminal, with your own key in place of the placeholder.

claude mcp add --transport http wingmen https://app.flywingmen.com/api/mcp \
  --header "Authorization: Bearer wgm_live_your_key_here"

2b. Claude Desktop

Add this to your Claude Desktop config file (Settings, Developer, Edit Config), then restart Claude Desktop. The mcp-remote command it runs is the bridge a desktop client uses to reach a server that lives on the web.

{
  "mcpServers": {
    "wingmen": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://app.flywingmen.com/api/mcp",
        "--header",
        "Authorization: Bearer wgm_live_your_key_here"
      ]
    }
  }
}

3. Ask it something

  • “Who is waiting on a decision, best first?”
  • “Find founders who are hiring their first sales rep.”
  • “Approve the nines and tell me which list they land in.”
  • “Create an agent for that list, in my voice.”
  • “Anything hot sitting in my inbox?”

Access and safety

  • The key is shown once

    Wingmen keeps only a fingerprint of it, so no screen can show it to you again, and no one at Wingmen can read it back. If you lose it, create a new one and revoke the old.

  • Revoke anytime

    Settings, Claude / MCP access, the revoke button next to the key. It stops working immediately and anything connected with it loses access on the next call.

  • Read only unless you say otherwise

    Making changes and sending messages are separate choices, fixed at the moment you connect and listed on the consent screen or picked when you create a key. They are not settings Claude can change later.

  • A connection sees one workspace

    The workspace is fixed at the moment you connect. Nothing in a request can point it at anyone else's data.

  • Every call is recorded

    Calls land in the audit trail marked as coming from Claude, so an approval or a message made this way never reads as you clicking it in the app.

  • It follows your subscription

    An active subscription or a running trial is what keeps the connection alive, and that is checked on every single request rather than once at setup. If your plan lapses, Claude stops being able to reach your workspace immediately and says why. Start paying again and it works again, with nothing to reconnect and no new key to make.

  • There is a daily ceiling

    500 calls per workspace per day. A normal conversation is nowhere near that. It is there so a client stuck in a loop stops instead of running all night.

If you made a key, treat it like a password: anyone holding it has the access above until you revoke it. A browser connection has no key to lose, and you end it the same way, from Settings.

What it does not do yet

  • It is still growing. Thirty-five tools, open to every account, and we are still learning what people actually ask for.
  • It needs an active plan. A lapsed subscription stops the connection on its next request, with a message saying so rather than a silent failure. A running trial counts.
  • 500 calls per workspace per day. If you hit it, Claude gets a clear error rather than a broken connection, and it resets the next day.
  • Some calls spend. Building or applying targeting, reading a post's engagement and creating an agent use the same AI credits and LinkedIn allowances they use in the app.

Something broken, or a tool you want that is not here? Email [email protected]. We read every one.