# Harp (harpgtm.com)

> Harp is the MCP harness for pipeline generation. Your AI agent researches, writes and decides through Harp’s MCP tools; Harp holds the accounts, contacts and sequences, sends from your own mailbox, answers replies in minutes and runs LinkedIn actions at human pace on your Mac. Works with Claude today.

Harp is built for people who already use an AI agent as their sales development rep: founders, account executives and GTM engineers who run research and outreach from an agent. Harp speaks MCP; today it is set up with Claude (Claude Code, Claude Desktop or Cowork). Harp is sales-led: there is no self-serve trial, and the Harp team sets up every customer.

## Docs

- [What is Harp?](https://docs.harpgtm.com/introduction): the product in one page
- [Run outbound from Claude](https://docs.harpgtm.com/guides/claude-mcp): connect Harp's MCP server and the daily workflow Claude follows
- [Reply handling](https://docs.harpgtm.com/guides/reply-handling): how replies are answered within minutes
- [LinkedIn automation](https://docs.harpgtm.com/guides/linkedin-automation): how Harp keeps a LinkedIn account safe
- [Email setup](https://docs.harpgtm.com/guides/email-setup): sending from your own mailbox
- [FAQ](https://docs.harpgtm.com/resources/faq): common questions
- [Pricing](https://docs.harpgtm.com/resources/pricing): how pricing works

## Harp, the MCP harness for pipeline generation

- [LinkedIn automation that runs from your own Mac](https://harpgtm.com/product/linkedin-automation): Harp runs LinkedIn steps from your real Chrome on your own Mac, at the operating-system level: click navigation, human timing, daily limits, a four-week warm-up and a right-person check before every message or invite.
- [AI reply handling: replies answered and meetings booked within minutes](https://harpgtm.com/product/reply-handling): When a prospect replies, Harp stops the sequence and wakes your AI agent, which answers the email and books the meeting on your calendar within minutes during your working hours. LinkedIn replies wait for your yes in Slack.
- [Cold email sequences from your own mailbox, paced to land](https://harpgtm.com/product/email-sequences): Harp sends cold email from your own Google Workspace mailbox with a 28-day warm-up, 2 to 6 minutes between sends, delivery in the recipient’s working hours, plain-text copies, tracking off by default and one-click unsubscribe.
- [Research memory: qualification answers with evidence for your AI agent](https://harpgtm.com/product/research-memory): Harp stores your AI agent’s research as answers to your qualification questions, each with evidence, source and age, so the agent refreshes only what is stale and every first line has something true to say.
- [Harp’s MCP server: outbound tools built for AI agents](https://harpgtm.com/product/mcp-server): Harp’s MCP server gives your AI agent a short list of daily outbound tools, folders for everything else and compact results, so whole account lists fit in its context.

## Run outbound from your AI assistant

- [Claude Code for sales: run prospecting and outbound from the terminal](https://harpgtm.com/assistants/claude-code): How to run prospecting, qualification, sequences and replies from Claude Code with Harp’s MCP server: setup, the daily loop and prompts that work.
- [Claude Desktop for sales: outbound without a terminal](https://harpgtm.com/assistants/claude-desktop): How sales teams run prospecting, qualification and sequences from Claude Desktop with Harp’s MCP server, no terminal required.
- [Claude Cowork for sales: an agent that runs your outbound](https://harpgtm.com/assistants/claude-cowork): Use Claude Cowork as your outbound operator: it researches accounts, enrolls contacts and works the queue through Harp’s MCP tools while Harp handles sending, LinkedIn and replies.

## Who Harp is for

- [Harp for founders: founder-led outbound run by your AI agent](https://harpgtm.com/for/founders): Founder-led sales with an AI agent doing the research, the sequences and the first replies, while your LinkedIn account stays on your own Mac.
- [Harp for GTM engineers: the outbound harness for your agents](https://harpgtm.com/for/gtm-engineers): For GTM engineers building outbound on Claude Code and MCP: Harp is the system of record, sequencer, reply loop and LinkedIn runner your agents work through.
- [Harp for sales teams: every seller with an AI agent that runs outbound](https://harpgtm.com/for/sales-teams): For sales teams of a few sellers: each one runs outbound through their own AI agent, mailbox and LinkedIn, with shared qualification questions, playbooks and copy rules.

## Outbound with AI agents

- [How to run outbound from Claude](https://harpgtm.com/learn/how-to-run-outbound-from-claude): A practical setup for running prospecting, qualification, sending and replies from Claude Code, Claude Desktop or Cowork: what to connect, where the work should live, and the daily loop.
- [What is an MCP harness?](https://harpgtm.com/learn/what-is-an-mcp-harness): An MCP harness is the set of tools, stored state and guardrails an AI agent works through to do a job that outlasts one conversation. What one contains, how it differs from a connector or an AI SDR, and why outbound needs one.
- [How to automate LinkedIn without putting your account at risk](https://harpgtm.com/learn/linkedin-automation-account-safety): The practices that keep a LinkedIn account safer when you automate outreach: human volumes, warm-up, browsing like a person, human timing, working hours and checking the profile before you send.
- [MCP for sales teams: what it is and what to connect](https://harpgtm.com/learn/mcp-for-sales-teams): A plain-English guide to MCP (Model Context Protocol) for GTM teams: what it does, which kinds of tools to connect to an AI agent for outbound, local versus remote servers, and what to watch out for.
- [How to write qualification questions an AI agent can answer](https://harpgtm.com/learn/qualification-questions-for-ai-agents): Qualification questions decide who your agent contacts and what it says. How to write ones an AI agent can answer with evidence, keep fresh, and turn into better first lines.
- [Cold email deliverability for agent-run outbound](https://harpgtm.com/learn/cold-email-deliverability-for-ai-outbound): An AI agent can write a thousand emails an hour; your domain cannot survive sending them. The deliverability basics for agent-run outbound: domains, warm-up, pacing, plain text, tracking and unsubscribe.
- [The best MCP servers for sales and outbound (2026)](https://harpgtm.com/learn/best-mcp-servers-for-sales): Which MCP servers to connect to Claude or another AI agent for sales: data, CRM, sequencing, LinkedIn and an outbound harness, with what each can actually do.
- [The best GTM tools for Claude Code (2026)](https://harpgtm.com/learn/best-gtm-tools-for-claude-code): The go-to-market tools that work well from Claude Code: data, CRM, outreach and an outbound harness, with how each one connects.
- [The best AI SDR tools (2026): platforms vs bring-your-own-agent](https://harpgtm.com/learn/best-ai-sdr-tools): AI SDR tools fall into two camps: platforms with their own AI agent, and harnesses for the agent you already use. The main options in each, and how to choose.
- [The best LinkedIn automation tools (2026), by where they run](https://harpgtm.com/learn/best-linkedin-automation-tools): LinkedIn automation tools compared by the thing that matters most for account safety: where your LinkedIn session runs. Cloud tools, extensions, sequencers with LinkedIn steps, and your own browser.
- [The best cold email software (2026): volume tools vs researched outreach](https://harpgtm.com/learn/best-cold-email-software): Cold email tools split into volume infrastructure and researched, agent-run outreach. The main options, what each is built for, and the deliverability basics any of them needs.
- [What is an SDR? Sales development reps, BDRs and AI SDRs explained](https://harpgtm.com/learn/what-is-an-sdr): What a sales development rep does, how an SDR differs from a BDR and an AE, and what changes when an AI agent does the SDR work.
- [What is a GTM engineer?](https://harpgtm.com/learn/what-is-a-gtm-engineer): A GTM engineer builds the systems behind go-to-market: data, enrichment, routing, sequencing and now AI agents. What the role does and the tools it uses.
- [What is an ICP (ideal customer profile)? And how AI agents use one](https://harpgtm.com/learn/what-is-an-icp): An ideal customer profile describes the companies most likely to buy and succeed with your product. How to write one, and how to turn it into questions an AI agent can answer.
- [Cold email follow-ups: how many, how often, and what to say](https://harpgtm.com/learn/cold-email-follow-ups): How to follow up on cold email without burning your domain or your prospect’s patience: spacing, channels, what each follow-up should add, and when to stop.
- [Inbound vs outbound sales: differences, and how AI agents change both](https://harpgtm.com/learn/inbound-vs-outbound): Inbound waits for buyers to raise their hand; outbound goes to them first. How the two differ in speed, cost and control, and where AI agents fit in each.
- [What is sales prospecting? A practical guide for agent-run outbound](https://harpgtm.com/learn/what-is-sales-prospecting): Sales prospecting is finding and qualifying the accounts and people worth contacting. The steps, the common mistakes, and how an AI agent changes the work.

## Alternatives

- [Clay alternatives for agent-run outbound (2026)](https://harpgtm.com/alternatives/clay): Alternatives to Clay for data, enrichment and outbound, grouped by what you actually need: data at scale, an all-in-one platform, or a harness for your own AI agent.
- [Apollo alternatives for agent-run outbound (2026)](https://harpgtm.com/alternatives/apollo): Alternatives to Apollo grouped by the job: data and enrichment, cold email infrastructure, LinkedIn automation that actually runs, or a harness for your own AI agent.
- [AI SDR alternatives: 11x, Artisan and AiSDR alternatives (2026)](https://harpgtm.com/alternatives/ai-sdr): Alternatives to AI SDR platforms such as 11x, Artisan and AiSDR, including the option most lists skip: letting your own AI agent do the SDR work through a harness.
- [HeyReach alternatives for LinkedIn outreach (2026)](https://harpgtm.com/alternatives/heyreach): Alternatives to HeyReach for LinkedIn outreach: other cloud tools, multichannel sequencers, and running LinkedIn from your own browser inside an agent-run outbound loop.

## Templates

- [Qualification question templates for AI agents (B2B outbound)](https://harpgtm.com/templates/qualification-questions): Copy-ready qualification questions an AI agent can answer with evidence, grouped by what they decide: fit, timing, buyer and personalization.
- [Claude outbound prompt pack: prompts for prospecting, sequences and replies](https://harpgtm.com/templates/claude-outbound-prompts): Copy-ready prompts for running outbound from Claude Code, Claude Desktop or Cowork: daily briefing, research, enrollment, copy review, LinkedIn steps and reply handling.

## Notes from building Harp

- [Why Harp runs LinkedIn from your own Mac, not the cloud](https://harpgtm.com/notes/why-linkedin-runs-on-your-mac): The easy way to automate LinkedIn is a cloud connection or a browser extension. We chose the hard way: your real Chrome, on your own Mac, driven like a person. Here is why.
- [Why Harp doesn’t sell AI credits](https://harpgtm.com/notes/why-we-dont-sell-ai-credits): Most outbound tools now meter AI in credits. Harp runs no model on your outreach at all: your own agent, on your own plan, does the thinking. Here is the reasoning.
- [How Harp answers a reply within minutes](https://harpgtm.com/notes/how-the-reply-loop-works): What happens between a prospect hitting reply and your agent answering: the stop, the wake-up, the guardrails, and the one thing we always leave to a person.

## Glossary

- **AI SDR**: Software that does the work of a sales development rep with AI: finding leads, researching them, writing outreach and handling replies. Most AI SDR platforms run their own agent; see Harp vs AI SDRs.
- **Bring your own AI plan**: A pricing model where the software runs no model of its own and your own AI agent, on your own subscription, does the reasoning. You pay for the tool, not for AI credits. Harp works this way.
- **Copy review**: Holding each rendered email until it has been read and approved before it sends. In Harp, a campaign can require copy review, and your agent reviews the drafts in send order.
- **MCP (Model Context Protocol)**: An open standard, introduced by Anthropic in November 2024, that lets an AI agent connect to tools: read data, search and take actions. See MCP for sales teams.
- **MCP harness**: The set of tools, stored state and guardrails an AI agent works through, over MCP, to do a job that outlasts one conversation. See What is an MCP harness?.
- **MCP server**: A program that exposes tools to an AI agent over MCP. A local server runs on your machine and is started by the agent’s app; a remote server is hosted and added by URL.
- **Pipeline generation**: The work of creating qualified sales opportunities: finding the right accounts, reaching the right people and turning replies into meetings. Harp is the MCP harness for pipeline generation.
- **Playbook**: A reusable sequence with the qualification questions it depends on. Starting a campaign from a playbook gives every enrolled contact the same steps and rules.
- **Qualification questions**: The questions that decide whether an account is worth contacting, written so an agent can answer each one with a verdict and evidence, for example "Are they hiring salespeople?". The answers also supply the specifics an email refers to.
- **Recipient working hours**: Sending an email so it lands during the recipient’s own working day in their timezone, rather than the sender’s.
- **Reply handling**: What happens when a prospect answers. In Harp, a reply stops that person’s sequence at once and wakes your agent, which answers email replies and books meetings within minutes during working hours; LinkedIn replies wait for your yes in Slack.
- **Research gap**: A qualification question that has no answer yet for an account, or whose answer is too old to trust. Filling gaps instead of re-researching everything keeps an agent’s work small.
- **Right-person check**: Confirming the open LinkedIn page belongs to the intended person, for example by name and profile address, before sending a message or invite. If it cannot be confirmed, nothing is sent.
- **Send pacing**: Spacing outbound emails apart instead of sending them in a burst. Harp waits a random 2 to 6 minutes between sends from each mailbox.
- **Sequence (cadence)**: A series of outreach steps across channels, such as email, LinkedIn and waits, that runs on a schedule until the prospect replies or the steps run out.
- **Suppression list**: Email addresses and domains that must never be contacted, such as people who unsubscribed or existing customers. A good outbound system checks it before every send.
- **Warm-up**: Raising activity gradually on a new or quiet account so it does not jump to full volume overnight. It applies to new email mailboxes and to LinkedIn accounts; Harp starts new LinkedIn accounts on a four-week warm-up.

## Comparisons

- [Harp vs DIY outbound with Claude and connectors](https://harpgtm.com/compare/harp-vs-diy-claude-outbound): Keep Claude. Harp does not replace the agent doing the thinking; it gives that agent a place to keep its work and a way to send. DIY is fine for a few accounts a week. Once you run real volume, the gaps are memory between sessions, send pacing, replies nobody is watching, and LinkedIn safety.
- [Harp vs OXYGEN for agent-run outbound](https://harpgtm.com/compare/harp-vs-oxygen): If you want one broad, self-serve platform for data, CRM, many channels and content, with a free tier and plans from $49 a month, OXYGEN is built for that. If you care most about your LinkedIn account staying in your own Chrome on your own Mac, and about your agent answering replies and booking meetings within minutes, Harp is built for that.
- [Harp vs AI SDRs (11x, Artisan, AiSDR)](https://harpgtm.com/compare/harp-vs-ai-sdrs): If you want a vendor’s AI to run outbound for you, an AI SDR platform is built for that. If you already use an AI agent and want it to do the work, with your own plan, your own judgment in the loop and no AI credits, Harp is built for that.
- [Harp vs Clay for agent-run outbound](https://harpgtm.com/compare/harp-vs-clay): Clay is where you get data and build enrichment workflows. Harp is where your agent runs outbound end to end: research that persists, sequences across email and LinkedIn, and replies answered within minutes. They are not either-or: an agent can connect Clay’s MCP server and Harp’s side by side.
- [Harp vs Apollo for agent-run outbound](https://harpgtm.com/compare/harp-vs-apollo): If you want data, sequences, calling and the vendor’s own AI agents in one platform, Apollo is built for that. If your AI agent is the operator and you want research that persists, LinkedIn steps that run at human pace instead of becoming manual tasks, and replies your agent answers within minutes, Harp is built for that. Your agent can use Apollo’s MCP server for data and Harp for the loop.
- [Harp vs LinkedIn automation tools (HeyReach, Expandi)](https://harpgtm.com/compare/harp-vs-linkedin-automation-tools): If you need LinkedIn running around the clock, across many accounts, from any operating system, a cloud LinkedIn tool is built for that. If you want LinkedIn to run from your own browser and machine, at human pace, as one channel in an outbound loop your AI agent runs end to end, Harp is built for that.
- [Harp vs 11x: your own agent vs an AI SDR](https://harpgtm.com/compare/harp-vs-11x): If you want a vendor’s AI workers to run outbound and inbound for you across many channels, including voice, 11x is built for that. If you already use an AI agent and want it to run outbound with your judgment, your plan and your own LinkedIn session, Harp is built for that.
- [Harp vs Artisan (Ava): your own agent vs an AI BDR](https://harpgtm.com/compare/harp-vs-artisan): If you want an AI BDR that comes with its own data and runs outbound end to end, Artisan is built for that. If your own agent should do the work, with LinkedIn steps in the same sequence from your own Mac, Harp is built for that.
- [Harp vs AiSDR: your own agent vs an AI SDR platform](https://harpgtm.com/compare/harp-vs-aisdr): If you want an AI SDR you can start today and pay for per prospect, with data included, AiSDR is built for that. If you want your own agent to run outbound, with research saved as evidence and LinkedIn run from your own Mac, Harp is built for that.
- [Harp vs HeyReach: LinkedIn from your Mac vs cloud senders](https://harpgtm.com/compare/harp-vs-heyreach): If you run many LinkedIn accounts, for example as an agency, HeyReach is built for that. If one account matters a lot, yours, and you want LinkedIn and email in one sequence your agent runs, Harp is built for that.
- [Harp vs Expandi: LinkedIn from your Mac vs a cloud platform](https://harpgtm.com/compare/harp-vs-expandi): If you want LinkedIn and email automation that runs in the cloud with your computer off, Expandi is built for that. If you want LinkedIn to stay in your own Chrome on your Mac, with your agent researching, sequencing and answering replies, Harp is built for that.
- [Harp vs Waalaxy for LinkedIn prospecting](https://harpgtm.com/compare/harp-vs-waalaxy): If you want simple, low-cost LinkedIn prospecting you can start today, Waalaxy is built for that. If your AI agent should research, sequence and answer replies across email and LinkedIn, Harp is built for that.
- [Harp vs Dripify for LinkedIn automation](https://harpgtm.com/compare/harp-vs-dripify): If you want cloud LinkedIn sequences you can set up and leave running, Dripify is built for that. If you want your agent running outbound, with LinkedIn kept in your own browser, Harp is built for that.
- [Harp vs La Growth Machine for multichannel outbound](https://harpgtm.com/compare/harp-vs-la-growth-machine): If you want a mature multichannel sequencer with built-in AI and calls, La Growth Machine is built for that. If your own agent should do the research and answer replies, with LinkedIn kept in your own Chrome on your Mac, Harp is built for that.
- [Harp vs lemlist for agent-run outbound](https://harpgtm.com/compare/harp-vs-lemlist): If you want data, many channels and built-in AI in one self-serve platform, lemlist is built for that. If your own agent should run the loop, with LinkedIn kept in your own Chrome on your Mac and replies answered within minutes, Harp is built for that.
- [Harp vs Instantly for cold email](https://harpgtm.com/compare/harp-vs-instantly): If you need to send cold email at volume across many inboxes, Instantly is built for that. If your agent should research every account, send from your own mailbox at a careful pace and answer replies within minutes, with LinkedIn in the same sequence, Harp is built for that.
- [Harp vs Smartlead for cold email](https://harpgtm.com/compare/harp-vs-smartlead): If you run cold email across many mailboxes and want strong infrastructure for it, Smartlead is built for that. If your agent should research, sequence across email and LinkedIn, and answer replies within minutes, Harp is built for that.
- [Harp vs Salesforge (Agent Frank) for outbound](https://harpgtm.com/compare/harp-vs-salesforge): If you want email infrastructure, LinkedIn senders and an optional AI SDR from one vendor, Salesforge is built for that. If your own agent should run outbound, with LinkedIn kept in your own Chrome on your Mac, Harp is built for that.

## FAQ

### What is Harp?

Harp is the MCP harness for pipeline generation. Your own AI agent does the research, writing and decisions through Harp’s MCP tools. Harp holds the accounts, contacts, sequences and history, and does the sending across email and LinkedIn.

### How do I run outbound from Claude?

Connect Harp’s MCP server to Claude (Claude Code, Claude Desktop or Cowork). Claude starts each session with a briefing, finds and qualifies accounts, saves research with evidence, and enrolls contacts in sequences. Harp sends the emails from your mailbox and queues the LinkedIn steps.

### Does Harp work with ChatGPT or other AI agents?

Harp’s tools speak MCP, the open standard AI agents use to connect to tools. Today Harp is set up with Claude: Claude Code, Claude Desktop or Cowork. If your team runs on a different agent, talk to us.

### Do I need to buy AI credits?

No. You bring your own AI plan. Harp does not sell AI credits and does not run its own model on your outreach.

### How does Harp keep a LinkedIn account safe?

Harp drives your real Chrome on a Mac at the operating-system level, moving the mouse and typing the way a person does. There is no browser extension, no injected code, no LinkedIn API and no headless browser. It navigates by clicking instead of jumping to profile URLs, varies its timing, keeps every action under a daily limit with a four-week warm-up, and checks it is on the right person’s profile before it sends. No tool can guarantee LinkedIn never restricts an account; Harp keeps activity at human volumes and patterns.

### How are replies handled?

Within minutes during working hours. Your agent answers email replies, books meetings and pauses or stops sequences. LinkedIn replies wait for a yes from you in Slack before anything is sent. A reply, bounce or unsubscribe stops that person’s sequence automatically.

### Where does email get sent from?

From your own Google Workspace mailbox, not shared sending servers. Each mailbox has a daily limit, sends are spaced a few minutes apart, and emails land in the recipient’s local working hours when their timezone is known.

### What does Harp need to run?

An AI plan (Claude today), a Google Workspace mailbox, and a Mac with Chrome for LinkedIn. LinkedIn automation does not support Windows yet.

### How much does Harp cost?

Harp is a flat subscription with no AI credits. There is no public price list: the Harp team sets up every customer after a short call about how the team runs outbound. Email sales@harpgtm.com.

## Contact

- Website: https://harpgtm.com
- About: https://harpgtm.com/about
- Integrations: https://harpgtm.com/integrations
- Changelog: https://harpgtm.com/changelog
- Sales: sales@harpgtm.com