YoYoGTM · Built in-house

We stopped renting an AI SDR. We built our own.

YoYoGTM finds the right people, warms them up on LinkedIn, writes to them one by one, follows up by email and hands every reply back for a human answer. It runs our own outbound every day, from a dashboard, a REST API, or straight inside Claude.

AI SDRLinkedIn + emailMCP serverREST APIBuilt with Claude Code
YoYoGTM overview: people in sequences, invites, acceptance rate, replies and meetings
Works inside Claude14-tool MCP server
The build
1 engine
finds, warms, writes, follows up and routes replies
99
buying signals in the catalog
14
tools in the MCP server
26
REST API routes, OpenAPI
0
per-seat fees
01

Why we built it instead of buying one

AI SDR tools charge per seat, per sender or per contact, and the good ones keep your data and your playbook inside their walls. We run outbound for ourselves and for clients, across several sending accounts. Paying a seat fee for every one of them, every month, for a tool we could not change, made no sense for a team that builds software for a living.

One AI SDR seat, rented for a year$99 × 12 = $1,188 per seat, before you add a second sender
ToolPlanPer monthPer year
GojiberryAI SDR, LinkedInPro$99$1,188
PumpGTMAI SDR via MCPStarter, per LinkedIn account$75$900
HeyReachLinkedIn automationGrowth, per sender$79$948
lemlistEmail + LinkedIn sequencesMultichannel, per user$109$1,308
11x (Alice)Autonomous AI SDRGrowth (reported)~$3,750~$45,000
YoYoGTMour own, on our own serverLinkedIn access via Unipile, up to 10 accounts~$55~$660

Vendor list prices checked on 13 November 2026 from each vendor's pricing page (11x is a reported figure; it does not publish prices). Our figure is the one fixed software cost we pay. AI writing and data lookups are pay-as-you-go on top, and the server is one we already run for other tools.

No seat tax

Add a sender, a campaign or a teammate without a new line on the invoice. Five sending accounts cost the same as one.

Our playbook, our data

Every person, message, reply and result sits in our own database. Nothing is locked in someone else's dashboard when we leave.

Rules we actually want

Verified-only email, high-income markets only, a do-not-contact list checked before every step. Off-the-shelf tools could not do all three.

Ships in an afternoon

When we want a feature, we write it. Sender pools, AI-picked calls to action and bounce detection all went live in days.

02

A complete outbound product, not a script

YoYoGTM is a full product: a web app, a sending engine, a review queue, a reply inbox, an API and an MCP server. It is the same kind of system we build for clients, running on our own pipeline first.

YoYoGTM overview dashboard

The overview

One screen for the whole funnel: people in sequences, LinkedIn invites, acceptance rate, replies and meetings booked, with a live activity feed of every visit, like, invite and email.

Choosing buying signals for a campaign

Signals, not spray

Campaigns start from buying signals: a new leader, a hiring spree, a competitor shipping an MCP server, a funding round. 99 signals in the catalog, each mapped to where the data comes from.

ClayLinkedInResearch agents
Reviewing people before they enter a sequence

A human review queue

Every person found is scored against the ideal customer profile with a reason. Approve or reject in one click, or let autopilot decide above a score you set.

The multichannel sequence editor

LinkedIn-first sequences

A visual sequence editor: profile visit, like a post, connect, message, then email two days later if there is no reply. AI writes the intro, hook and call to action for each person, and you preview exactly what will send.

Claude writingPer-lead variables
Answering a reply from the replies inbox

Replies come back to you

Every reply lands in one inbox with a drafted answer. Nothing is ever sent on a reply without a person pressing send.

Running YoYoGTM from inside Claude

Run it from Claude

The 14-tool MCP server means you can plan a campaign, review people and answer replies by talking to Claude. It signs in with OAuth, the way Claude's connector directory expects.

MCPOAuth 2.1

Screenshots use demo data. Every name, company and email in them is a sample.

03

The engine behind every message

Outbound fails on the boring parts: deliverability, pacing and bad data. That is where most of the engineering went.

0%bounce target: email only goes to verified addresses
5sending accounts in one pool, auto-balanced
30emails per account per day, hard cap
2 daysbefore the email follow-up, LinkedIn first
  1. FindPlain-English campaign brief in, people out. Search and enrichment run through Clay.
  2. ScoreClaude scores fit against your ICP and explains why. Famous or oversized companies are marked down.
  3. VerifyEmails are checked by ZeroBounce, then Findymail for catch-alls. No verified address, no email.
  4. Warm upProfile visit and a like before the invite, paced inside daily LinkedIn limits.
  5. WriteOne AI writing pass per person per step, cached, so the preview is exactly what sends.
  6. SendEach person gets the least-busy inbox in the pool and keeps it for the whole thread.
  7. Route repliesLinkedIn and Gmail replies are polled, bounces caught, meetings matched from Cal.com.
04

How we built it

Senior judgment plus Claude Code. The first working version, from empty folder to live on our server, took days, not months.

Architecture

One registry, three surfaces

Every action is defined once. The MCP server, the REST API and the dashboard all call the same 14 operations, so they can never drift apart.

Safety

Sandbox by default

Every new workspace starts in sandbox mode. Nothing reaches a real person until the owner turns it off, and every write tool is marked as destructive for Claude.

Deliverability

Bounces caught everywhere

Gmail files bounce notices in their own threads, so a scanner checks every mailbox every 30 minutes and re-checks right before each follow-up.

Platform

API, MCP, OAuth, plugin

A public REST API with an OpenAPI spec, idempotency keys, revocable API keys, an OAuth 2.1 server for Claude and a Claude Code plugin.

Claude CodeTypeScriptNode 22Postgres 17MCP SDKUnipileGmail APIClayZeroBounceCal.comSlackCoolify
05

Hiring someone to build outbound? This is the proof.

These are the kinds of projects we are asked to build every week. YoYoGTM is a working answer to each of them, already running.

AI outboundBuild and run our AI outbound engine, LinkedIn plus cold email

This is that engine: LinkedIn steps, email follow-ups, pacing and reply handling in one system.

Signal-basedClay + n8n LinkedIn outreach automation with signal-based personalization

Signals pick the people, Clay enriches them, and Claude writes a first line from what it found.

Inside your appBuild a cold email engine inside our product, with a sending platform API

YoYoGTM is a product with an API. We can build the same engine into yours.

GTM engineeringGTM engineer to automate prospecting, scoring and follow-up

Find, score, verify, write and route: every step here is automated and logged.

AI agents + MCPBuild an AI agent our team can drive from Claude

Our MCP server lets a team run outbound by talking to Claude, with human approval on anything that sends.

DeliverabilityCold email infrastructure with near-zero bounces

Verified-only sending, sender pools and bounce scanning are built in, not bolted on.

Your turn

Want an outbound engine you own?

Tell us how you sell today. In 30 minutes we will map the smallest version of this system that would pay for itself, on your tools and your data.

Free 30-minute call. Honest take, even if that means “you don't need us yet.”

Roadmap call · 30 min · free

Loading the calendar… If it does not appear, book on cal.com.

Want an AI SDR you own?· Free 30-min call →