Grammarly checks how you write. Assay checks what your AI says about your company. The wrong price, spelled perfectly. A claim legal never approved, in flawless English. Caught before it ships, right inside ChatGPT, Claude, and Cursor.
Draft the security paragraph for the Meridian proposal.
Meridian data is isolated per tenant✓ P-014, encrypted with AES-256 at rest and TLS 1.3 in transit✓ P-022, and hosted in your choice of EU or US region✓ P-031. Customer data is deleted within 30 days of termination✓ P-047.
Zero workflow change
The workflow change is zero. Not small. Zero.
Your people keep working where they already work. Assay connects as an MCP server, your AI reads the house rules at connect, and the same governed answer shows up in every host, sealed.
"Meridian data is deleted within 30 days of termination, in your EU or US region."
✓Assay seal · verify"Meridian data is deleted within 30 days of termination, in your EU or US region."
✓Assay seal · verify"Meridian data is deleted within 30 days of termination, in your EU or US region."
✓Assay seal · verifyThe idea engine
Someone has a sharp GTM thought inside a private AI chat, and it's gone by lunch. Assay notices, asks, and, only if they say yes, files it, credited to them.
"What if we led the Meridian deck with the 30-day deletion guarantee instead of burying it in the appendix? It's the thing their security team actually asked about."
Keep it? It'll be credited to you and queued for testing.
The best positioning ideas your company has are already being spoken, once, into a chat window. This is the net under them.
Nothing is captured without an explicit yes. The author's name rides with the idea into testing, so credit lands where it's due.
The Assay Seal
A hallmark didn't promise the ring would make you happy. It certified the metal was what the seller said it was, and who stood behind the claim. The Assay Seal is that, for what your AI says about you.
"Customer data is deleted within 30 days of termination."
Every governed answer carries a cryptographic receipt of exactly which claims were checked, against which version of your truth.
Anyone who sees the Seal can open the receipt. The check is inspectable, not a word you're asked to take.
The Seal appears only on a real check that passed. No check, no checkmark. An unconfigured setup plainly says unsigned.
Work in progress carries a draft mark, not a Seal, so a half-checked answer can never masquerade as a sealed one.
The Seal is a record of the check, never a promise the answer is right.
That distinction is the whole ethic. A hallmark certifies the metal, not the buyer's taste. The Seal certifies the claim was checked against your approved truth, and gives anyone the receipt to confirm it. When nothing was checked, it says so.
House rules + access
The moment a host connects, it learns what it may read, what it may write, and how to ask. Access travels with the seat, and every write is a deliberate, two-step action.
/assay-pulse for what changed, /assay-context to ground a draft, /assay-check to verify a claim. Present the moment your AI connects.
A connection can only ever read and write what that person's seat is permitted to. Revoke the seat, revoke the reach.
Changing the source of truth is never a side effect. It's proposed, then confirmed by a permitted human, then logged.
One surface, seven agents
Connect once and the team is reachable from inside your AI. Here's what each does through the MCP today, stated honestly.
Grounds and seals answers about your category. Live in the MCP today.
Available nowSurfaces what's winning and what changed via /assay-pulse.
Early accessRuns /assay-check on any claim and mints the Seal.
Early accessGrounds posts in your voice and your approved truth.
Early accessDrafts from live nodes so nothing ships stale.
Early accessAnswers a rep's question on today's story, sealed.
PreviewChecks outbound copy against the gate before it runs.
PreviewStatus is what each agent does through the MCP today, not a roadmap. The chips mean exactly what they say.
Straight answers
RAG retrieves text and hopes the model uses it. Assay grounds the answer in your governed, approved truth, checks each claim against it, and attaches a signed receipt for the ones that pass. Retrieval is a step. Governance with a verifiable receipt is the product.
No. Assay checks what your AI says about your company, not what your people type. The idea engine only captures something when a person explicitly consents, credited to them and logged. It offers, you decide.
No. It runs inside the tools your people already use. The workflow change is zero, not small. Zero. Your AI reads the house rules at connect and the checks are just on.
The Seal is earned pixels only. It appears on a governed answer that passed a real check, never on an empty result or an unconfigured setup, which plainly says unsigned. No check, no checkmark.
No. The Seal is a signed receipt with a public verify link. It records that a specific check ran against a specific version of your truth. It's a record of the check, never a promise the answer is right.
You can connect the SEO & GEO agent and start free. The MCP connection reads your approved truth and mints the Seal on governed answers. Writing back to the source of truth is a governed, two-step action reserved to seats with permission.
No. Assay produces readiness evidence, an audit log, and the artifacts a reviewer asks for. It doesn't certify you compliant, because no software can. Compliance is your determination, made with your counsel.
Connect your AI
Add Assay as an MCP server in the host you already use. It reads your approved truth, mints the Seal on governed answers, and reads the house rules at connect.
# Claude · Settings → Connectors → Add MCP server { "mcpServers": { "assay": { "url": "https://mcp.assay.com/sse", "headers": { "Authorization": "Bearer <your-seat-token>" } } } }
Prefer to see what the engines say about you first? Run the audit →