The Sharpmatter Protocol

How AI gets to actually answer ad questions.

Three deliberate stages, one continuous protocol. Connect every channel once. Compress hundreds of thousands of rows into a tight token stream. Then ask Claude anything — without it hallucinating.

01Connect

Plug in every client's channels. Once. Read-only. Revocable.

Each client's accounts live in an isolated workspace with its own instructions. OAuth into the platforms your clients actually run — Sharpmatter unifies every campaign, click, conversion, and cent into a single private surface per client. We never write back, never sell, never share.

me

Meta Ads

Campaigns, ad sets, ads, insights

go

Google Ads

Search · Performance Max · YouTube

ti

TikTok Ads

Spark Ads · TopView · Shop

ga

Google Analytics 4

Sessions, conversions, funnels

st

Stripe

True revenue · MRR · refunds

mi

Microsoft Ads

Bing search · LinkedIn audiences

Pinterest, LinkedIn, Reddit, and Snapchat launching in the coming months. Custom integrations on Enterprise.

02Compress

Hundreds of thousands of rows. One AI-shaped signal.

The compression layer is the moat. Without it, AI hits its context window in seconds and starts hallucinating. With it, Claude can hold your full account in a single conversation — losslessly on what matters.

Step 01

Fetch

Sharpmatter calls each ad platform's API in parallel. ~147,000 rows of raw JSON come back per query.

Step 02

Normalize

Every platform's schema gets translated to a single canonical format. Meta's 'spend' becomes the same as Google's 'cost' and Stripe's 'gross'.

Step 03

Deduplicate

Multi-touch attribution overlap is collapsed. Cross-platform halo effects are quantified, not double-counted.

Step 04

Compress

We strip every field the model doesn't need. ~147,000 rows become ~482 tokens of structured signal — losslessly on the meaning.

Raw rows fetched

147,382

Tokens delivered to AI

482

Compression ratio

99.67%
03In your Claude

Your team asks. In the Claude they already use.

Once your data is compressed, Claude sees the signal — not the noise. Ask in Claude Desktop, claude.ai, Cursor, or any MCP client — switch clients by asking, and Sharpmatter serves compressed, isolated, per-client data underneath.

Its own instructions

Each client workspace carries its own compliance rules, targets, and tone — Claude reads them before it answers.

No cross-client bleed

Data, context, and conversation history stay scoped to the workspace you're in. Nothing leaks between clients.

Switch by asking

Tell Claude to move to another client and it does — no separate login, no separate app.

Unlimited internal seats

Your whole team works inside the same client workspaces at no extra cost.

Try it live

See it in action.
Get started.

Priced by client count — from $49/mo.