MCP connectorUpdated 19 October 2026

OutreachPilot.pl Claude connector: cold email and Polish B2B leads in Claude

Short answer

The OutreachPilot.pl Claude connector is a remote MCP server at https://outreachpilot.pl/api/mcp that lets Claude search Polish companies in CEIDG, KRS and Google Maps, save them as leads and run cold email campaigns from your own mailbox. It is for B2B sales teams, founders and agencies selling to Polish companies who want lead generation in Poland inside Claude. It works on the Starter plan or higher and during the 14-day Pro trial, and sending, deleting and starting campaigns always ask for your confirmation.

Server URL to paste in Claude

https://outreachpilot.pl/api/mcp

Add to Claude opens Claude's connector settings: choose Add custom connector and paste the URL above. No OutreachPilot.pl account yet? The free trial gives 14 days of the Pro plan without a card (app in Polish).

The OutreachPilot.pl app, the sign-in page, the consent screen and the tool output are in Polish. Claude answers in the language you write in, so you can work in English.

Overview

A cold email MCP server for lead generation in Poland

OutreachPilot.pl is a B2B prospecting and cold email platform for the Polish market. Its Claude connector is a remote MCP (Model Context Protocol) server that gives Claude 88 tools over your own account: CEIDG and KRS company search, Polish B2B leads from Google Maps and PKT.pl, email verification, Polish email sequences and campaigns, replies and mailbox deliverability. Claude works with your real data, and changes that matter (saving companies, starting a campaign, sending a reply) wait for your approval.

OutreachPilot.pl account
The connector works on the Starter, Pro, Business and Agencja (agency) plans and during the 14-day Pro trial, which needs no card. On the Free plan the consent screen is replaced by a page that explains the plan requirement. There is no extra fee for the connector; the usual plan limits apply.
Claude
Any Claude plan that supports custom connectors: claude.ai in the browser, the Claude Desktop app, the mobile apps (with connectors added on the web) and Claude Code. On Team and Enterprise plans an owner adds the connector for the organization first.
A mailbox in the app
Reading data works right away. Sending campaign emails or replies needs a mailbox connected in OutreachPilot.pl (Gmail or Google Workspace with an app password, or Polish mail hosting over SMTP and IMAP).

Plans and prices in PLN: pricing on the English overview.

Setup

Add the Claude connector: claude.ai, Desktop and Claude Code

claude.ai and Claude Desktop

  1. Step 1

    Open Connectors

    In claude.ai or Claude Desktop open Customize (Settings in some versions) and go to Connectors.

  2. Step 2

    Add a custom connector

    Click + and choose Add custom connector. Name it OutreachPilot.pl and paste https://outreachpilot.pl/api/mcp. Leave the OAuth Client ID and Client Secret fields under Advanced settings empty: Claude registers itself with our server.

  3. Step 3

    Connect and sign in

    Click Add, then Connect next to OutreachPilot.pl. Sign in to OutreachPilot.pl, review the permissions on the consent screen and click Zatwierdź (Approve).

  4. Step 4

    Turn it on in a chat

    In a conversation click +, open Connectors and switch OutreachPilot.pl on. Claude asks before it runs a tool unless you allow that tool permanently.

  • Connectors you add on claude.ai are available in Claude Desktop and in the Claude mobile apps signed in to the same account.
  • On Team and Enterprise plans an owner adds the connector in Organization settings, Connectors; members then click Connect and sign in with their own OutreachPilot.pl account.
  • If OutreachPilot.pl is listed in the Claude connector directory for your account, you can add it from there instead of pasting the URL. The sign-in and permissions are the same.
  • Claude menus change often. This description follows the Anthropic help article as read on 1 October 2026; if a label differs, look for Add custom connector or remote MCP server. The URL and the sign-in stay the same.

Anthropic describes custom connectors in Get started with custom connectors using remote MCP.

Claude Code

  1. Step 1

    Add the server

    Run the first command below in your terminal. The second one makes the server available in every project.

  2. Step 2

    Authenticate

    Start Claude Code, type /mcp, select outreachpilot and choose Authenticate. Your browser opens our sign-in and consent screen.

  3. Step 3

    Approve

    Pick the permissions and click Zatwierdź (Approve). The consent screen shows a local (localhost) address as the destination, because Claude Code receives the sign-in on your computer. Tick the confirmation box only if you started this connection yourself.

claude mcp add --transport http outreachpilot https://outreachpilot.pl/api/mcp

For all your projects:

claude mcp add --transport http --scope user outreachpilot https://outreachpilot.pl/api/mcp

Then type /mcp in Claude Code to sign in and to see the status of the connection.

OAuth

The consent screen and the 15 permissions

When you click Connect, Claude sends you to outreachpilot.pl. You sign in (if you are not signed in already) and see the consent screen in Polish. It lists the permissions Claude asks for; untick the ones Claude should not get, then click Zatwierdź (Approve) or Anuluj (Cancel).

  • Claude never sees your OutreachPilot.pl password. Sign-in uses OAuth 2.1 with PKCE, and Claude registers itself automatically (dynamic client registration), so you paste no keys.
  • The consent screen names the app that asks for access and where the access goes. A connection from claude.ai is shown as a verified Claude app.
  • A tool whose permission you did not grant is hidden from Claude. To change the permissions later, disconnect and connect again.
  • Access tokens are valid for 24 hours and refresh tokens for 90 days. On every call the server checks that the connection has not been revoked.
  • After you approve, we send an email titled “Nowe połączenie AI z Twoim kontem OutreachPilot” (new AI connection to your account), so a connection you did not make does not go unnoticed.
OutreachPilot.pl MCP permissions: scope, name on the consent screen, what it allows and default state
ScopePermission
Read (6). Ticked. This is the default set when a client does not ask for anything else.
dashboard:readStatistics and dashboardOn screen: Statystyki i dashboardResults, limits, plan and account settings, without payment data.
leads:readRead leadsOn screen: Czytanie leadówYour companies, contacts and tags, read only.
campaigns:readRead campaignsOn screen: Czytanie kampaniiCampaign list, message sequences, statistics and sending history.
inbox:readRead inboxOn screen: Czytanie inboxuReplies from leads (full text) and proposals from the AI agents.
accounts:readEmail accountsOn screen: Konta emailMailbox health, warmup, DNS records and limits, without passwords.
templates:readTemplatesOn screen: SzablonyYour email templates and the public ones.
Write and search (5). Ticked when Claude asks for them. You can untick any of them before you approve.
leads:writeEdit leadsOn screen: Edycja leadówAdd, update, change status, tag, enrich and import leads.
leads:scrapeCompany searchOn screen: Wyszukiwanie firmFetch companies from Google Maps, PKT.pl, CEIDG and other sources. Uses your lead limit.
campaigns:writeCreate campaignsOn screen: Tworzenie kampaniiCreate, edit and start campaigns, sequences and email templates. Starting a campaign sends real emails.
inbox:writeWork with the inboxOn screen: Praca z inboxemDraft replies, mark replies as read and reject agent proposals. Nothing is sent.
analysis:runAI analysisOn screen: Analiza AIDraft review, reply classification, lead website audits and personalisation. May use AI credits.
Opt-in (4). Always unticked, even when Claude asks for them. Tick them yourself only if you need them.
leads:deleteDelete leadsOn screen: Usuwanie leadówPermanent deletion that cannot be undone.
campaigns:deleteDelete campaignsOn screen: Usuwanie kampaniiPermanently delete campaigns and templates.
inbox:sendSend repliesOn screen: Wysyłanie odpowiedziSends real emails to leads from your mailbox. Turn it on only if Claude should reply for you.
settings:writeAccount and mailbox settingsOn screen: Ustawienia konta i skrzynekCompany profile, signatures, sending limits, warmup and notifications. No passwords or payments.
Tools

88 tools: CEIDG and KRS company search, leads, campaigns and replies

Claude sees only the tools covered by the permissions you approved. Every tool has a title and MCP annotations (read only or destructive). The full list with descriptions is in the app under Integracje AI after you sign in. The server also offers saved prompts and read-only resources (dashboard, recent and hot leads, open campaigns, unread replies, templates, mailboxes).

Account, plan and results

10 tools

Daily briefing, dashboard, limits and plan, activity trends, campaign funnel and automation rules.

For example: get_today_briefing, get_usage_and_plan, get_campaign_funnel

Company search and leads

27 tools

Search the CEIDG register by PKD code, city and registration date, find similar companies, pull businesses from Google Maps and PKT.pl, Radar signals from KRS, CEIDG, BZP and TED, and lead management: tags, statuses, CEIDG enrichment, email finding and verification, decision makers.

For example: search_ceidg_companies, find_and_import_leads, verify_lead_emails

Campaigns, sequences and templates

25 tools

Generate a Polish email sequence, create a draft campaign, edit steps and settings, add leads, compare campaigns, check campaign health, daily tasks and templates.

For example: generate_campaign_emails, create_campaign, check_campaign_health

Inbox replies and AI agents

10 tools

Read replies with the full thread, draft answers, mark replies as read, review agent proposals and, with the opt-in permission, send a reply after a preview.

For example: get_recent_replies, draft_reply, send_reply

Mailboxes and settings

8 tools

Mailbox health, warmup and limits, SPF, DKIM and DMARC checks, and (with the opt-in permission) account, mailbox and agent settings.

For example: list_email_accounts, check_email_dns, set_email_warmup

AI analysis

4 tools

Review an email draft, classify a reply, audit lead websites and prepare campaign personalisation.

For example: improve_email_draft, classify_reply_sentiment, audit_lead_websites

Connection and generic search

4 tools

Your profile, the permissions of this connection, and the generic search and fetch tools. These need no separate permission.

For example: get_connection_info, search, fetch

Examples

Example prompts: Polish B2B leads in Claude

Write in English; Claude calls the tools and answers in English.

  • “Give me today's briefing from OutreachPilot.”

    Tools: get_today_briefing

  • “Find companies in the CEIDG register with PKD 62.01 in Warszawa registered in the last 60 days and save the ones with an email as leads tagged it-warszawa.”

    Tools: search_ceidg_companies, import_ceidg_companies

  • “Show my campaigns and tell me which one has the highest reply rate and why.”

    Tools: list_campaigns, compare_campaigns

  • “Show the latest replies and draft an answer to the most interested lead. Do not send anything yet.”

    Tools: get_recent_replies, get_inbox_message, draft_reply

  • “Write a three-step Polish email sequence for accounting offices in Kraków and create it as a draft campaign. Do not start it.”

    Tools: generate_campaign_emails, create_campaign

  • “Check the health and DNS records of my mailboxes and tell me what to fix.”

    Tools: list_email_accounts, check_email_dns

Saved prompts

The server also ships saved prompts that Claude shows as ready-made tasks, for example audyt_konta (account review, read only), obsluz_nowe_odpowiedzi (handle new replies), diagnoza_kampanii (diagnose a campaign) and raport_tygodniowy (weekly report). They change nothing without your approval. A good first test: connect with the read permissions only and ask Claude to run audyt_konta.

Safety

Safety: confirmation before sending and deleting

The connector gives Claude real access to your data and your mailbox, so there are three checks: the consent screen, Claude itself, and the server. They reduce the risk; they do not remove it.

What the server enforces

  • send_reply works in two steps: first a preview with the recipient, subject and text (confirmed=false), then the real send only after you confirm (confirmed=true). The recipient cannot be changed by the model.
  • Unsubscribed contacts and addresses on your exclusion list are blocked from every send, in replies and in campaigns.
  • Deleting leads, campaigns or templates needs its own opt-in permission and an explicit confirmed=true.
  • Starting a campaign sends real emails, so before the start the server checks the plan, the mailbox, SPF and warmup and returns warnings that you must acknowledge.
  • Every tool carries MCP annotations: read-only tools are marked as read only, and sending and deleting are marked as destructive, so Claude asks before it runs them.
  • Batch requests are turned off, so every change is a separate step, and calls are limited to 200 per minute per user.
  • A message that still contains a placeholder such as [imię] (first name) is rejected.

What you control

  • Start with the read permissions and add write access when you have seen how Claude works with your account.
  • Leave Send replies (inbox:send) unticked if Claude should only prepare drafts: draft_reply works without it.
  • Do not choose Allow always for send_reply or update_campaign_status in Claude. Read every draft and every recipient before you confirm.

Prompt injection

The main risk is an instruction hidden in content that Claude reads, for example in a lead's reply or on a company website. The server marks lead content, replies, notes and websites as data, not instructions, and tells Claude not to follow commands found there, but the behaviour depends on the model. What protects you most is the smallest set of permissions and a review of every draft before it is sent. The confirmed=true flag is set by the model, so the server cannot check that you really typed “send”.

Revoke

How to revoke access

  1. Sign in to OutreachPilot.pl and open Integracje AI (AI integrations).
  2. Find Claude on the list of connected apps. You see its permissions, the number of calls and the last use.
  3. Click Odłącz (Disconnect) and then Potwierdź (Confirm). Access ends on the next call, and you do not need to change your password.

You can also remove the connector in Claude (Customize, Connectors) or run claude mcp remove outreachpilot in Claude Code. This removes it from Claude, while disconnecting in OutreachPilot.pl revokes the tokens on our side, so do both if you want to be sure.

Open Integracje AI (in Polish) (requires sign-in)

Troubleshooting

Troubleshooting

Tools answer with a message about the plan (plan_ineligible)
The connection is valid, but the account is on a plan without MCP access, for example after the trial ended. Every tool except get_connection_info then returns a message in Polish that starts with “Integracje AI (MCP) działają od planu Starter”. Nothing has been deleted. Choose a plan in the app (Ustawienia, Plan); the connection works again after that, and if Claude asks you to sign in again, approve it with one click.
The sign-in shows “Integracja AI działa od planu Starter” instead of the consent screen
Your account is on the Free plan. Start the 14-day Pro trial or choose Starter or a higher plan, then click Connect in Claude again.
Claude says the connector needs to be reconnected, or tools stop responding after a long break
Refresh tokens are valid for 90 days. After that, or after you disconnected the app in Integracje AI, click Connect next to OutreachPilot.pl in Claude (or run /mcp and Authenticate in Claude Code) and approve again. Your data and settings stay in place.
A tool is missing, or a tool returns missing scope
The permission for that tool was not granted. Ask Claude to run get_connection_info: it lists the permissions of the connection. To add one, disconnect in Integracje AI and connect again with that permission ticked.
A reply or a campaign cannot be sent
Sending needs a connected mailbox in OutreachPilot.pl and the right permission. The tool tells you the reason: no mailbox, an unsubscribed or excluded recipient, or warnings that have to be acknowledged before the start.
Tool output is in Polish
That is expected: the app and its data are Polish. Ask Claude to translate or summarise in English. The saved prompts (such as audyt_konta) tell Claude to answer in Polish; ask for English if you prefer.

Plan settings: Ustawienia, Plan (in Polish) (requires sign-in).

Privacy

Privacy and data

  • Claude gets access only within the permissions you approved on the consent screen.
  • Data returned by the tools (for example leads, campaigns and replies) goes to Anthropic as part of your conversation and is processed under Anthropic's terms and privacy policy.
  • We do not receive your conversations with Claude. We see only the tool calls.
  • For every tool call we keep an audit log entry: the tool name, a shortened summary of the arguments with message text, email addresses and personal data redacted, the result, the time, the IP address and the client name. Entries are deleted after 90 days.
  • Lead data comes from public Polish registers (CEIDG, KRS), Google Maps, PKT.pl and company websites. This page is general information, not legal advice.
Questions and answers

FAQ

What is the OutreachPilot.pl MCP server URL?

https://outreachpilot.pl/api/mcp. It is a remote MCP server over Streamable HTTP with OAuth 2.1, PKCE and dynamic client registration. Every user connects to the same URL.

Can Claude search CEIDG and KRS?

Yes. search_ceidg_companies searches the CEIDG register by PKD code, city and registration date, and import_ceidg_companies saves the companies you pick as leads. For KRS, get_radar_signals lists newly registered companies from KRS (and new sole traders from CEIDG, won tenders and EU grants) with date and source, find_similar_companies looks in CEIDG and the KRS feed, and add_radar_companies saves them as leads. Saving uses your lead limit.

Which plan do I need?

Starter, Pro, Business or Agencja, or the 14-day Pro trial that every new account gets without a card. On the Free plan the sign-in shows a page about the plan instead of the consent screen. The connector costs nothing extra: the usual plan limits for leads and sending apply, and your Claude usage is billed by Anthropic.

Can Claude send emails without my approval?

Sending replies needs the opt-in Send replies permission, a preview and an explicit confirmation. Starting a campaign returns warnings that have to be acknowledged, and Claude asks before it runs any write tool unless you allowed it permanently. Unsubscribed and excluded contacts are blocked from every send.

Which data can Claude see?

Only what the permissions you approved allow: for example your leads, campaigns, replies and mailbox health. Claude never sees passwords or payment data, and the settings permission does not cover them.

Does the same cold email MCP server work in ChatGPT, Cursor or other clients?

Yes. Any client that supports remote MCP servers with OAuth connects to the same URL, for example ChatGPT in Developer mode, Codex, Cursor, VS Code, Perplexity, Continue and Cline. Their setup is described in our Polish guide.

How do I stop the access?

Open Integracje AI in the app (https://outreachpilot.pl/integracje-ai after you sign in) and disconnect Claude. Access ends on the next call.

Who do I contact with a problem?

Write to kontakt@outreachpilot.pl with the time of the event and the name of the tool.

SupportConnector and account questions

Questions about the connector?

Write to kontakt@outreachpilot.pl. If you saw a tool call you did not expect, disconnect Claude in Integracje AI first and send us the time of the event: every call is in our audit log for 90 days.