The brain, as a server.

1,626 campaigns of proven pattern, wired into every agent you run - self-hosted, governed, and yours to build on.

1,626 campaigns encoded - the graph every product below reasons from.

Live

Ask the graph. It answers in a few tool calls.

The knowledge graph is nodes and edges - segments, angles, copy patterns, objections, outcomes - each tier-scored by how much human-approved proof stands behind it.

$ claude -> gtm-mcp (self-hosted)

mcp query "what converts for Series A RevOps?"

✓ graph.search_segments

✓ graph.match_angles

✓ graph.rank_copy_patterns

✓ returned: winning angles, segments, copy patterns

Illustrative query shape. Tool-call counts, timing, and match totals vary per query; see PROOF NEEDED below.

PROOF NEEDED - a real captured query (tool-call count, reply rate, matching-campaign count) to replace the illustrative example above. The demo figures previously shown here (110+ live GTM sources, 3 tool calls not 33 / 91% token saving, 92 matching campaigns · 4.1% avg reply) had no source anywhere in the workspace and were removed.

What it is

The brain every other product reasons from.

A self-hosted, governed GTM knowledge graph served over the Model Context Protocol. Every answer is assembled from reviewed claims, carries citations back to exact evidence, and states its own limits - nothing is generated from a model's recollection.

1,626 campaigns, encoded

Real campaign outcomes turned into a queryable pattern store - what converts, for whom, and why - not a pile of raw notes. The canonical store holds 715 imported rows (665 tools, 32 campaigns, 11 offers, 7 personas) plus 464 documents

- source: tools/mcps/gtm-mcp/README.md.

Speaks MCP

Any MCP-capable agent - Claude and beyond - queries it as tool calls. The brain plugs into your agents, not the other way around.

Self-hosted, multiplayer

Two transports - local stdio for an editor or agent on your machine, and a protected, subscriber-gated Cloudflare Worker for production.

Three-tier trust

Emergent to validated to ground truth - only human-approved facts graduate, so the brain compounds without going gullible.

Governed, not guessed

A canonical D1 store, immutable R2 snapshots, and two independent gates on every gated answer - authentication, then a fail-closed subscriber check. Fetching is allowlist-first: six trusted-source registry rows today, four eligible, two gated

- source: tools/mcps/gtm-mcp/README.md.

Thirteen tools, one registry

Seven legacy lookup tools plus six governed answer products, each returning citations, trust tier, freshness, and contradictions in one envelope.

How your agents use it

Query the graph. Reason from proven pattern.

The graph holds segments, angles, copy patterns, objections, and outcomes - each scored by how much human-approved proof stands behind it.

Auth into the server

Your agent connects to the self-hosted MCP server and licenses access to the graph, scoped to you.

Ask in plain language

"What converts for Series A RevOps?" resolves against 1,626 campaigns in a few tool calls, not a research session.

Get pattern, not opinions

Winning angles, segments, and copy patterns come back grounded in real reply and revenue outcomes.

Feed your wins back

Your audit data writes into your own graph. Validated facts graduate to ground truth and the moat compounds.

Part of the operating system

Everything that reasons from it.

GTM OrchestratorReads from the brain to run campaigns.
Fast ReplyWrites replies from proven copy - page coming.
Mail LegionPrivate managed email infrastructure, exclusive to members.

A thousand soldiers. One general. You.

Borrow 1,626 campaigns of pattern from day one.

GTM MCP comes with every founding seat - the brain your whole operating system reasons from, and the graph you build your own moat on top of.

Join the waitlist ->