MCP · Connect your AI agent

Run outbound from the AI you already use.

Connect Claude, ChatGPT, Codex, Cursor or any MCP client with one URL. Your agent sets up the account, plans campaigns, writes or reviews messages and handles replies, while you approve what goes out.

Connect your assistantSee the 50 tools
  • Claude
  • ChatGPT
  • Claude Code
  • Codex
  • Cursor
  • VS Code
  • any MCP client
Illustration: a laptop with chat bubbles sends a glowing cable to a workshop island of research, writing and mail machines.
One URL carries your assistant into research, writing and replies. It can draft and plan. Messages go out only through campaigns you approved.streamable HTTP · OAuth 2.1

Connect

Pick your assistant. Paste one URL.

Most clients sign in with OAuth from the URL alone. The rest use an API key instead.

Claude on web, desktop and mobile

Claude adds custom connectors from a link, so the URL is already filled in.

  1. Click Add to Claude. It opens the custom connector form with the name and URL set.
  2. Click Add, then Connect.
  3. Approve Agentic Outreach on our consent page. The account owner approves.
  4. Say Set up Agentic Outreach for me.
Add to Claude

Opens Claude in your browser.

On Team or Enterprise, an org owner adds the connector first. Anthropic decides which plans allow custom connectors, so check your plan.

  1. Add the URL
  2. Approve on the consent page (account owner)
  3. Say: Set up Agentic Outreach for me

Whether a Claude or ChatGPT plan allows custom connectors is decided by Anthropic and OpenAI. Check your plan. MCP access itself is included on every Agentic Outreach plan.

Agent-led onboarding

Your agent sets up the account. You answer a few questions.

After you connect, paste one line: Set up Agentic Outreach for me. The dashboard checklist fills in live while it works. The whole setup takes about 10 minutes.

  1. Reads what it already knows

    It starts from its memory of you and from your website, so you don’t re-enter what it can already see.

  2. Shows one summary to confirm

    One draft of your company, offer and ideal customer. You confirm it or correct it.

  3. Asks only the gaps

    Two or three questions at a time, three to six in total.

  4. Saves the profile, ICP and messaging

    Your company profile, your ideal customer profile with a free audience estimate from the database, and your messaging rules.

  5. Hands you a connect link

    A link to connect LinkedIn or your email sender. You open it, and the agent checks progress.

  6. Launches the first campaign

    Prospects are enrolled from the database for free. Nothing is sent until you approve Day 1.

Illustrative conversation

Your assistantagentic-outreach
  • Set up Agentic Outreach for me.
  • get_onboarding_status→ 3 of 6 steps open · next: confirm profile
  • analyze_website→ draft profile · not saved
  • From your site, you sell fraud screening to payments teams at mid-market companies. I’d target heads of risk and VPs of payments in the US and UK. Is that right?
  • Yes. US and UK only.
  • save_icp→ 412 matching decision-makers already in the database · free
  • Saved. Next, your messaging rules, then a link to connect LinkedIn.
  • launch_first_campaign→ nothing sent · Day 1 awaits your approval
  • The first campaign is ready. Approve Day 1 from the dashboard when the messages look right.

Dashboard checklist, filling in live

  • Company
  • Offer
  • ICP
  • Messaging
  • Senders
  • First campaign

Day to day

Then it runs the loop with you.

Type what you want in plain language. Each prompt triggers a short chain of tools, and nothing reaches a prospect until you say yes.

  • Say this

    “Who’s warm this week? Draft openers for the best two.”

    • find_warm_leads
    • get_lead_evidence
    • draft_outreach

    Previews only. Nothing is sent or saved.

  • Say this

    “Plan a campaign for CFOs at Series B fintechs in the UK.”

    • plan_campaign
    • preview_audience
    • create_campaign

    Audience research is free. Creating a campaign sends nothing.

  • Say this

    “Write the messages yourself from the research.”

    • list_pending_messages
    • get_message_context
    • write_messages

    Used in client mode, where your agent writes the copy.

  • Say this

    “Draft replies to anyone who answered.”

    • list_pending_replies
    • write_reply_draft

    Drafts land in the inbox panel. You send them with one click.

  • Say this

    “Show me tomorrow’s queue.”

    • get_schedule
    • approve_day

    approve_day runs only after you say yes.

  • Say this

    “How are campaigns doing?”

    • campaign_insights

    Counts replies, meetings and sends by channel. Opens are not tracked.

AI mode

You choose who writes the copy.

One setting for the whole account. Either Agentic Outreach writes every message, or your connected agent does.

server mode

Agentic Outreach writes

Our writer drafts each first touch and follow-up from the research. You can edit any message before launch.

  • Writes from each prospect’s research and signals.
  • Follows the messaging rules you saved.
  • Day 1 still waits for your approval.
client mode

Your agent writes

Your connected agent reads each research dossier and writes the messages itself, with write_messages.

  • Reads each dossier with get_dossier.
  • Saves up to 100 messages per call with write_messages.
  • Nothing is written until your agent writes it. Follow-ups wait in list_pending_messages.

You can switch either way later with set_ai_mode. Your agent confirms with you first.

Playbooks

Five playbooks, read before the work starts.

Each playbook is the routine for one kind of work. Your agent reads it before starting, so it works the way the playbook says.

  • onboarding

    Sets up a new account from what your agent already knows, plus a short interview for the gaps.

  • copywriting

    Writes first touches and follow-ups from each prospect’s research, inside the platform’s rules.

  • launch-campaign

    Turns a goal into an ICP, sizes the audience for free, picks channels and gets Day 1 ready for approval.

  • daily-operator

    Runs the daily routine: status sweep, today’s messages, the send batch for approval, and campaigns that underperform.

  • reply-handling

    Triages replies and drafts answers that move the conversation toward a meeting.

  • Three ways they reach your agent

    • MCP promptsClaude Code lists them as /mcp__agentic-outreach__<name>.
    • get_playbookCalled with a name, or with none to list them. Works in every client, ChatGPT included.
    • ZipA zip for native skill installs.

Tools

All 50 tools, by the job they do.

These are the tools your agent can call. Scopes (read, draft, campaign, approve) limit what each key can do.

50 tools

    • get_onboarding_statusThe setup checklist and the single next step. Start here for a new account.
    • get_setup_questionsThe onboarding question bank, with what is already answered.
    • analyze_websiteReads the company website and returns a draft profile. Nothing is saved yet.
    • save_company_profileSaves who the user is and what they sell. Partial updates are fine.
    • save_icpSaves the ideal customer profile and returns a free audience estimate.
    • save_messagingSaves tone and rules for outreach. They apply to every campaign.
    • get_connect_linkA link the user opens to connect a LinkedIn or email sending account.
    • launch_first_campaignCreates the first campaign from the saved setup. Nothing is sent until Day 1 is approved.
    • complete_onboardingMarks onboarding finished, so the full dashboard is available.
    • get_playbookReads one of the five playbooks. Call with no name to list them.
    • whoamiThe account, plan, connected senders, company profile and size of the prospect pool.
    • get_billing_statusPlan, credits left, trial days and when credits reset.
    • list_connected_accountsConnected email and LinkedIn sending accounts and their status.
    • list_industriesIndustry ids in the prospect pool, with how many prospects each holds.
    • preview_audienceFree audience research against the existing database, with a ranked sample.
    • plan_campaignGuided intake. Asks one question at a time until the campaign draft is complete.
    • plan_campaign_daysGives each unplanned prospect a channel and send day, within daily caps.
    • create_campaignCreates a campaign and fills it from the database first. Nothing is sent.
    • set_campaign_icpWrites a campaign's ideal customer profile, formed by the agent.
    • run_discovery_gapSources fresh prospects for the remaining gap. Spends credits, so only with your OK.
    • discovery_statusDiscovery progress and enrollment counts for one campaign.
    • crawl_company_listCrawls a public company directory into a clean list. No credits.
    • enroll_company_listFinds the right people at a list of companies you choose and enrolls them.
    • track_competitorWatches a competitor and turns the people who engage with them into warm leads.
    • list_competitor_trackersThe competitors being tracked, and the campaign their leads flow into.
    • list_campaign_prospectsProspects enrolled in a campaign, with contact and verification status.
    • research_prospectsRuns full research on a campaign's prospects: posts, news, funding, hiring, competitors.
    • get_dossierThe full research dossier for one prospect, with source-cited evidence.
    • draft_messageDrafts one message for a prospect from their dossier. Preview only.
    • list_pending_messagesThe copy queue: first touches and follow-ups still waiting for a message.
    • get_message_contextEverything needed to write a batch of messages in one call.
    • write_messagesSaves copy your agent wrote into a campaign, up to 100 items at once.
    • save_prospect_insightsStores your synthesis of a prospect's research on the account.
    • get_scheduleThe approval queue for one send day, with the exact subject and body of each message.
    • approve_dayApproves one send day so its messages go out. Only after you say yes.
    • pause_campaignStops a campaign's scheduled touches. Nothing is deleted.
    • resume_campaignResumes a paused campaign.
    • get_ai_modeShows who writes the copy: Agentic Outreach or your agent.
    • set_ai_modeSwitches who writes the copy. Confirm with the user first.
    • list_pending_repliesConversations waiting on a reply, newest first.
    • write_reply_draftSaves a reply as the pending draft in the inbox. Nothing is sent.
    • find_warm_leadsWarm leads surfaced by signal agents, filterable by status, kind and score.
    • get_lead_evidenceOne warm lead, the signals that surfaced it, and its score.
    • draft_outreachPreviews the first message for a warm lead. Nothing is sent or saved.
    • approve_warm_leadEnrolls an approved warm lead into a campaign. Needs the approve scope.
    • list_agentsSignal agents with their mode, status and rolling stats.
    • pause_agentPauses a signal agent. Needs the approve scope.
    • campaign_insightsCounts per campaign: enrolled, replied, meetings and sends by channel.
    • list_campaignsCampaigns with their status and enrollment counters.
    • get_campaignFull detail for one campaign: ICP, counters and discovery state.

Safe by construction

Your agent can draft. It has no send tool.

Four controls decide what an agent can touch. Each one is listed with what it does.

OAuth 2.1 with PKCE

Your client signs in from the URL alone. Access tokens last 24 hours. Refresh tokens last 180 days and rotate. Both are stored hashed. Disconnect any agent in one click from Settings → API & MCP.

Four scopes

Access is split into read, draft, campaign and approve. Approving a warm lead or pausing an agent needs the approve scope.

No send tool

There is no tool that sends a message. Messages go out only through campaigns you approved, inside daily limits. Approving a send day is the one step that contacts people, and your agent runs it only after you say yes.

Spend needs your OK

Audience research against your database is free. Fresh paid sourcing waits for your explicit OK, and credit use is capped by your plan.

Read the security pageWe have not completed a SOC 2 audit.

FAQ

Before you connect.

Do I need an API key?

Not for Claude, ChatGPT, Claude Code, Codex, Cursor or VS Code. They sign in with OAuth from the server URL alone. An API key is the fallback for clients without OAuth. You create one in Settings → API & MCP. It is shown once and can be revoked.

Which assistants work?

Any client that speaks MCP over streamable HTTP. This page has install steps for Claude on web, desktop and mobile, ChatGPT, Claude Code, Codex, Cursor and VS Code. Any other MCP client connects with the URL and an API key.

Does my agent see my data?

It sees your account only, and only what its scopes allow: read, draft, campaign or approve.

Can my agent send messages directly?

No. There is no send tool. Messages go out only through campaigns you approved, inside daily limits. A sequence stops when a prospect replies, bounces or opts out.

Can I disconnect?

Yes. The Disconnect button in Settings → API & MCP revokes that agent instantly. Access tokens last 24 hours. Refresh tokens last 180 days and rotate.

What if my client doesn’t support OAuth?

Create an API key in Settings → API & MCP and send it as a Bearer token. Keys start with ofk_, are shown once, are stored hashed and are limited to 120 requests per minute.

Is MCP access extra?

No. MCP and AI-agent access is included on every plan.

Do I need a paid Claude or ChatGPT plan?

Check your plan. Anthropic decides which Claude plans allow custom connectors, and OpenAI decides the same for ChatGPT. ChatGPT Developer mode needs a paid plan.

Who approves the connection?

The account owner. You approve on our consent page, and you can revoke the agent later from Settings → API & MCP.

Your agent is one URL away. Request access.

Private beta. Access is by waitlist, and billing is not live yet.

Request accessSee pricing