MCP platform · Metered tools · Signed receipts

The execution layer AI agents use to run GTM.

Humans buy outcomes. Agents call tools. Same machinery, same meter, same receipts. The MCP tools below are the same ones our own apps run on — find contacts, enrich them, send the sequence, handle the reply, book the meeting.

01 · What the agent knows, what the agent does

OrgDrive reads out. The tools write in.

Your agent doesn't need another database to talk to. It needs a place to look things up and a place to take action — and it needs both to be the same system, so that what it does today makes what it knows tomorrow sharper.

Reads out · OrgDrive

What the agent knows

  • Account and contact state — who's in play, who's gone quiet
  • Behavioral signals with the full engagement trace behind them
  • Transparent scores, with the reasoning a rep could defend
  • Campaign verdicts — what was killed, what earned the right to scale
Writes in · MCP tools

What the agent does

  • find_contacts · enrich_contact
  • compose_outreach · send_sequence
  • handle_reply · book_meeting
  • write_crm — every action and outcome, logged

Everything that writes into OrgDrive arrives through these tools. Everything the tools do gets smarter from OrgDrive. That's the whole loop — there is no third path, and no private back door.

02 · The tool surface

Seven verbs. Every one of them metered.

This is the entire write surface. Your agent calls these over MCP exactly the way our own apps call them — same endpoint, same meter, same ledger.

ToolWhat it doesStatus
find_contacts Build a target list from an ICP definition Live
enrich_contact Firmographic + behavioral enrichment on a contact or account Live
compose_outreach Personalized sequence drafts from enrichment + your value prop Live
send_sequence Multi-step, deliverability-managed sending. Warmed domains, throttling, bounce handling. Live
handle_reply Classifies inbound, drafts or sends the response, escalates on low confidence Beta
book_meeting Negotiates time, sends the invite (Ush3r) Beta
write_crm Every action + outcome logged to OnePgr CRM or pushed to Salesforce/HubSpot Live
MCP endpoint: https://chain.onepgr.com/mcp
Approval mode is the default. Full-auto is earned, per domain, with caps and a kill switch.
03 · The meter

Per-outcome is the price. Per-action is the floor.

Agents that can spend money need a meter their owner can read. Actions bill at cost-level rates. The thing you actually wanted — a meeting on a calendar — bills as an outcome.

Per action · the floor
Enrichmentenrich_contact
Sendsend_sequence, per message delivered
Reply handledhandle_reply, per classified inbound
Per outcome · the price
Meeting booked

A real meeting, on a real calendar, with a qualified contact who agreed to take it. Not a click. Not an intent score.

Per-outcome is the price. Per-action is the floor. You never pay for a meeting twice, and you never pay for machinery you didn't run.

04 · The action ledger

Every vendor hands your agent a score. We hand it the receipts.

Every action gets a signed, timestamped receipt — what was sent, to whom, what happened. Your billing record, your audit trail, your provenance. An agent that spends money on your behalf should be able to show its work.

receipt r_9d31c tool send_sequence actor agent:acme-sdr-bot contact c_84f21 action step 2 of 5 delivered timestamp 2026-07-14T09:41:22Z outcome delivered · opened metered 1 send signature sha256:8f2a…c710
Your billing record. The ledger is the meter. Every charge on your invoice resolves to a receipt with an action, a recipient, and an outcome — the same per-outcome math your CFO can model.
Your audit trail. When an agent sends on your behalf, RevOps can reconstruct exactly what it did and why. Provenance isn't a feature request here. It's the record the system is built on.
05 · No private APIs

Our own apps run on the same tools.

Every LLM-native app we ship is an MCP connector app calling the surface above. There's no faster internal path we kept for ourselves.

Our apps have no private APIs. If Kampaign can do it, your agent can do it.

06 · Guardrails

An agent with a send button needs a brake.

Autonomy is earned, domain by domain. Until it is, a human signs off — and the brake is always within reach.

Approval mode by default
Per-domain send caps
Suppression lists
Brand-voice constraints
One-click kill switch
Build on it

Your AI agent can talk about GTM. Point it at the tools that do it.

Bring your agent, your platform, or your RevOps team. We'll wire it to the MCP endpoint, set the caps, and turn on the meter. Approval mode first — full-auto when you've seen the receipts.