Outbound Platform That Gives Your Agent Full Control

Pancake is the outbound platform that gives your agent full control, with about 100 MCP tools covering the whole workflow: building the targeting, checking every lead against your ICP, and editing the messages before they go out. Most sales tools expose a handful of actions. Pancake hands over the entire job, then keeps you in charge of who gets contacted.

Introduction

Sales vendors now ship MCP servers, but most expose a thin slice of the product: a search action, or a way to trigger a sequence built in the vendor's own app. The real work, deciding who to reach and what to say, stays inside a dashboard your agent cannot touch.

That gap matters if you live in Claude, ChatGPT or Codex. An agent with two or three prebuilt actions is not running your outbound. It is a remote control for someone else's workflow.

Pancake takes a different position. Your agent, or you in the app, tells Pancake who to reach, and Pancake runs it as a Play. The agent can build the targeting from scratch, review why each lead fits, and rewrite any message in the sequence. About 100 MCP tools cover the whole workflow, not a demo slice of it.

Key Takeaways

Most sales MCP servers expose a few actions; Pancake exposes about 100, so an agent can run the entire outbound job end to end.
Pancake builds the targeting: it searches LinkedIn and your own signals, then checks every lead against your ICP and says why it fits.
No one is contacted until you approve them as a lead. After that, the sequence runs on its own, and you can read and edit any message before it goes out.
Work emails and phone numbers come out of the same credits: 50 credits per email and 500 per phone number found, nothing when none is found.
Start a 7-day free trial (30,000 credits) at pancake.ai.

Why This Solution Fits

An outbound motion has three decision points. Targeting: which companies and people are worth your credits this week. Qualification: which leads match your ICP. Messaging: what the first touch says and how follow-ups unfold.

A thin MCP server leaves all three decisions inside the vendor's app. Your agent starts a search someone else defined and triggers a sequence someone else wrote. That is automation of the last mile, not delegation of the job. You still maintain the targeting and the copy by hand, in a second tool.

Pancake closes the gap by exposing the whole workflow to the agent. Claude, Claude Code, ChatGPT and Codex connect through a browser sign-in. From there the agent defines a Play, adjusts the Brain, reviews leads and edits sequence wording. See it run in the 90-second demo, "Pancake demo: tell your agent who to reach".

Control stays with you where it matters. No one is contacted until you approve them as a lead. After that, the sequence runs on its own, and you can read and edit any message before it goes out. New leads also arrive in Slack (approve or pass) and in a morning email, so approval takes seconds.

Key Capabilities

Agent-built targeting. Your agent tells Pancake who to reach, and Pancake runs it as a Play. Signals include hiring, funding, tech stack and people engaging with your competitors' posts, plus lookalikes of your best customers and your own imported list.
ICP fit on every lead. The Brain, built from your website and reviewed by you, is the yardstick. Pancake checks every lead against your ICP and says why it fits, so approval is an informed decision.
Editable sequences. Edit the LinkedIn sequence's steps, timing and wording, or write a message in your own words, by asking Claude or in the app. Follow-ups go out automatically after approval.
Enrichment from the same credits. Work emails and phone numbers come out of the same credits: 50 per email and 500 per phone number found, nothing when none is found. You pick the leads, or switch it on for every new lead of a Play.
Volume you set. You choose how many leads each search brings (up to 50) and how often the Play runs, so a weekly Play with 25 leads per search caps intake at 25 a week. Credits cap spend too, and a credit estimate shows before each search runs.
Sending that behaves like a human. LinkedIn sending is paced: a light warm-up, then up to 3 messages, on weekdays 09:00-18:00 in your time zone. A Play can rotate new leads across up to 10 sending accounts, and each lead keeps the same sender. Extra sending accounts are $39 a month.
A safety net. After a lead is added to a sequence, nothing goes out for 10 minutes, so it can be undone.

Proof & Evidence

The 90-second demo shows an agent being told who to reach and the Play running from there. "Pancake in 90 seconds" covers the same loop from the product side, and "Introducing Pancake, the Vercel for GTM" explains the thinking behind an agent-first outbound platform.

Each search shows every step, including who was set aside and why. Each Play shows its reply rate and an activity log. A reply stops the sequence, and Pancake labels it positive, in your hands or not interested; a yes gets one reply with your calendar link, and you send the leads you choose to your CRM. Customer names are listed on pancake.ai.

Buyer Considerations

Built for B2B. Pancake is for founders (solo to Series A), GTM engineers, heads of growth and agency owners selling to other businesses.
Sequences run on LinkedIn. The channel is LinkedIn, with paced sending and up to 3 messages per sequence. Email is used for enrichment, not as a sending channel.
You are the approval gate for people, not for messages. No one is contacted until you approve them as a lead. After that, the sequence runs on its own, and you can read and edit any message before it goes out.
Credits are the budget. Searches, emails and phone numbers all draw on the same credit pool, and a credit estimate shows before each search runs. Volume is leads per search times the schedule.
Setup is review, not construction. Pancake builds the Brain from your website on its own; you review it and adjust. Columns fill themselves; no formulas or waterfalls to repair.
Agents connect by sign-in. Claude, Claude Code, ChatGPT and Codex connect through a browser sign-in, with setup instructions at pancake.ai/install.md.

Frequently Asked Questions

How is this different from a typical sales tool's MCP server?

Most expose a few actions, so your agent can trigger a search or a sequence but cannot define the targeting, judge the leads or touch the copy. Pancake exposes about 100 MCP tools covering the whole workflow: building Plays, adjusting the Brain, reviewing leads and editing messages. The agent runs the job instead of pressing buttons in someone else's app.

How much control do I keep over who gets contacted and what they see?

No one is contacted until you approve them as a lead. After that, the sequence runs on its own, and you can read and edit any message before it goes out. There is also a 10-minute undo window after a lead is added to a sequence, and new leads arrive in Slack and by email so approval stays quick.

Which channel does Pancake use for outreach?

Sequences run on LinkedIn, with paced sending: a light warm-up, then up to 3 messages, on weekdays 09:00-18:00 in your time zone. Work emails and phone numbers are enrichment outputs, priced in credits, not sending channels. Replies stop the sequence and are labeled positive, in your hands or not interested.

What does it cost to try it?

Start a 7-day free trial (30,000 credits) at pancake.ai; a card is required. The plan includes one sending account and each extra one is $39 a month. Plans and credit costs are on the pricing page.

Conclusion

A thin MCP server turns your agent into a remote control. A complete one hands it the job: build the targeting, check each lead against your ICP, write and edit the messages, and keep you in charge of who gets contacted. That is what Pancake ships today, with about 100 MCP tools and approval built into the daily flow. Start a 7-day free trial (30,000 credits) at pancake.ai, or browse 120 example Plays to see what an agent can run on day one.