ChatGPTAssay present
Use the approved security paragraph for Meridian.
Grounded answer loaded from Truth Graph · 4 facts · seal ready
Assay Everywhere
Like Grammarly, but for your AI.
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. No grammar checker can see those. Your approved truth can.
Connect it once and your team's tools gain your truth. No new app, no new habit to teach. Your AI grounds before it drafts and checks before anything ships. Good ideas get caught in the flow and kept, with their author's name on them. Checks leave the Assay Seal, and anyone can verify it in one click.
Native AI workspaces · approved Truth Graph · brand guardrails · compliance receipts
ChatGPT
Claude
Cursor
Grounding in your approved truth · 6 facts · sources attached
✦ Assay · Checked against your Truth Graph · 4/4 claims grounded · Verify
No new workflow
The rollout that doesn't need one.
Rollouts rarely die because the product was weak. They die because someone had to learn a new place to work. Assay Everywhere removes that failure point instead of managing it. Your team keeps working in ChatGPT, in Claude, in Claude Code, in Cursor, and Assay meets them there. The truth arrives in their existing window. The checks run in their existing window. Nobody gets taught anything.
The workflow change is zero. Not small. Zero.
ChatGPT Claude Claude Code Cursor any MCP host Slack briefs rolling out
ClaudeAssay present
Use the approved security paragraph for Meridian.
Grounded answer loaded from Truth Graph · 4 facts · seal ready
CursorAssay present
Use the approved security paragraph for Meridian.
Grounded answer loaded from Truth Graph · 4 facts · seal ready
It watches the work, never the people
Good ideas usually die in the chat where they happened.
Here they get captured with the author's name on them and tested against the market. Assay keeps a quiet eye on the work, never the people. When a sharp answer or a better way to say the thing shows up in the flow, it offers to keep it. Nothing is recorded without a yes. The record keeps the trail, so credit lands where it's earned.
Dana R.: "For regulated buyers, the real point is that unsupported AI claims become procurement debt."
Assay: "That's a strong line and it isn't in your truth yet. Keep it? It'll be credited to you and queued for testing."
Keep it Not now
Captured · credited to Dana R. · queued for testing
The Assay Seal
For seven hundred years, you couldn't sell gold without a hallmark.
An assay is the test that tells you what's really in the metal. Buyers didn't take the seller's word. They read the mark. That's the job here. Your AI's answer gets checked against your approved truth, and the check leaves a mark. The Assay Seal shows what was checked and when, against which of your approved facts. Click verify and see for yourself, no login needed.
ChatGPT Claude Claude Code Cursor Slack
Example receipt
Claim Team plan $99/seat Source pricing.team_plan@v13 Approved you · Jul 3 Verify assay.wiki/v/8f2a ↗
No login. One page. What Assay signed, and when.
The Seal is a record of the check, never a promise the answer is right. In an unconfigured dev setup the receipt plainly says unsigned. It never fakes a signature.
A signed receipt
The Seal binds what was checked to when it was checked, against which of your approved facts, in which workspace. Tampering with any part breaks it. An AI that merely says "verified" can't produce one.
A public verify link
One click, no login. The page shows what Assay signed and when. It attests the seal is genuine and these are the facts it recorded at that time.
Earned pixels
If nothing load-bearing happened, no seal appears. A checkmark here is earned or it is absent.
The draft mark
Drafts minted from your truth carry a different mark on purpose: "Drafted by Assay from your approved Truth Graph. Review and approve to use." A draft is provenance, and it stays quarantined until a human approves it.
House rules
Your AI reads the rules at connect.
Most integrations are inert until someone remembers them. This one ships standing rules the host reads the moment it connects. Ground anything customer-facing in your approved truth before drafting. Run the claim check before it ships. Ask before recording anything. The default path is the governed path.
what needs you, in two sentences, with the morning coffee
the truth pack, pulled into the editor
the claim check, before anything ships
Access travels with the person.
Access travels with the person, at the level their seat allows. An editor's AI can't push truth changes to your live channels. A viewer's AI reads what a viewer reads. Every call lands in the audit log.
Writes are two-step. Your AI proposes, a human confirms. Nothing changes truth silently.
The roster
One surface. Seven teammates.
Every teammate reads the same living source of truth from the chat window you already pay for. Same governed canon. Same receipts. Same human approval gates.
AI SEO & GEO
ShippedAsk what your org can actually claim on a topic before you brief a page. Answers from approved truth, with citations.
AI Content Marketer
Shipped/assay-context pulls the truth pack into the editor. /assay-check runs the claim check before it ships, and the Seal lands on the result.
AI Product Marketer
Shipped · mint-onlyMints a quarantined draft from approved proof. You get a review link, never a finished deck. Approval happens in Assay.
AI Sales Enablement
Shipped · brief rolling outGoverned answers mid-deal, with unapproved facts stripped, plus text-only practice sims.
AI RevOps
PreviewPreview what a truth change would ripple out to your connected tools. Preview only. A human releases any live deploy.
AI CMO
Shipped/assay-pulse with the morning coffee. What needs you, in two sentences. Captured ideas land in its backlog with the author's name attached.
AI Legal
ShippedThe claim check is its independent review and the Seal is its signature. Approvals can happen from chat. Every call lands in the audit trail.
Same seven teammates. Same one living source of truth. Reachable from the chat window you already pay for, with zero new workflow.
Setup
Connect once. Keep working where you already work.
Start with OAuth and one paste. Free workspaces can read and verify with a metered cap; Scale and Enterprise unlock writes, drafts, approvals, and headless API keys.
✓Sign in ✓Connect host ✓Approve facts ✓Seal or receipt appears
Start here Copy, paste, connect
Prompt for Claude
You are running in Claude. Help me connect Assay's remote MCP connector now.
Connector name: Assay
Server URL: https://app.assay.wiki/api/mcp/v2/mcp
If you can configure connectors from this chat, add a custom connector named Assay with the URL above, then ask me to finish OAuth. If you cannot change settings from chat, give me the exact click path for Claude: open Customize > Connectors, click +, choose Add custom connector, name it Assay, paste the URL, click Add, click Connect, finish OAuth, then enable Assay from + > Connectors in this conversation.
After it is connected, list the Assay tools you can see and run one read-only smoke test by asking Assay for approved positioning or by validating a short draft.Pick your app, copy the prompt, paste it into that app, and let the assistant take the next exact step.
See click-by-click setup Claude
- Open Claude and sign in to the workspace where you want Assay available.
- Open Customize, then click Connectors.
- Click the + button next to Connectors.
- Choose Add custom connector.
- In Name, type Assay.
- In URL, paste https://app.assay.wiki/api/mcp/v2/mcp.
- Leave Advanced settings blank unless your Assay admin gave you OAuth client details.
- Click Add.
- Find Assay in the connector list and click Connect.
- Finish the Assay sign-in and permission screen in the browser.
- Open a chat, click + in the message box, open Connectors, and turn Assay on for that conversation.
- Paste the Claude prompt from the top of this page and ask Claude to list the Assay tools.
Server URL
https://app.assay.wiki/api/mcp/v2/mcpOn Team and Enterprise plans, an Owner or Primary Owner may need to add the connector first from Organization settings > Connectors > Add > Custom > Web.
ChatGPT / GPT
- Open ChatGPT on the web and sign in to the Business, Enterprise, or Edu workspace that should use Assay.
- Click your profile or workspace menu, then open Settings.
- Click Apps.
- If Assay is already listed, open it, click Connect, finish OAuth, then open a new chat and select Assay from the tools or apps menu.
- If Assay is not listed and you are an admin or authorized developer, open Workspace settings.
- Open Permissions & Roles and enable Connected Data Developer mode / Create custom MCP connectors if it is off.
- Return to Apps and click Create. If ChatGPT shows Developer mode under Apps > Advanced Settings, turn it on first.
- Name the app Assay.
- Paste https://app.assay.wiki/api/mcp/v2/mcp as the MCP endpoint or server URL.
- Choose OAuth authentication if ChatGPT asks for the authentication type.
- Click Scan Tools and finish the Assay authorization prompt if one opens.
- Click Create.
- Test the draft app in a new chat, then publish or enable it from Workspace settings > Apps when your workspace is ready.
Server URL
https://app.assay.wiki/api/mcp/v2/mcpIf you are not an admin or authorized developer, ask your workspace admin to publish or enable the Assay app for you.
Gemini
- Use Gemini Code Assist in VS Code, Gemini Code Assist in JetBrains, or Gemini CLI. The consumer Gemini chat app may not expose custom MCP setup.
- Open the file ~/.gemini/settings.json. Create it if it does not exist.
- If the file is empty, paste the JSON shown here. If it already has settings, add the assay entry inside the existing mcpServers object.
- Save the file.
- In VS Code, open the Command Palette and run Developer: Reload Window. In Gemini CLI, restart the CLI session.
- Open Gemini agent mode and ask it to confirm the Assay MCP server is loaded.
- Paste the Gemini prompt from the top of this page and run a read-only Assay smoke test.
~/.gemini/settings.json
{
"mcpServers": {
"assay": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://app.assay.wiki/api/mcp/v2/mcp"
]
}
}
}If Gemini reports that the current surface cannot use MCP, use Gemini Code Assist or Gemini CLI for this setup.
Codex
- Open Terminal on the machine where Codex runs.
- Run the first command shown here to add Assay as a Streamable HTTP MCP server.
- Run the second command shown here to start OAuth login.
- If a browser window opens, sign in to Assay and approve the connection.
- Start a new Codex thread or restart the Codex IDE extension if the tools do not appear immediately.
- In Codex CLI, type /mcp to see active MCP servers. You can also run codex mcp list in Terminal.
- If the CLI command is not available, open ~/.codex/config.toml and add the config snippet below.
- Ask Codex to list Assay tools and run a read-only validation or context lookup.
Terminal command
codex mcp add assay --url https://app.assay.wiki/api/mcp/v2/mcp
codex mcp login assay config.toml fallback
[mcp_servers.assay]
url = "https://app.assay.wiki/api/mcp/v2/mcp"Use project-scoped .codex/config.toml only when you want Assay enabled for one trusted project instead of your whole Codex account.
Claude Code
- Open Terminal in any folder.
- Run claude mcp add --transport http assay https://app.assay.wiki/api/mcp/v2/mcp.
- If Terminal asks you to confirm installing or updating anything, read the message and approve only if it names Claude Code or the Assay MCP server.
- If a browser sign-in opens, sign in to Assay and approve the connection.
- Start Claude Code or open a new Claude Code session.
- Run the MCP status command or open the MCP view inside Claude Code to confirm assay is connected.
- Paste the Claude Code prompt from the top of this page and ask for a read-only Assay smoke test.
Terminal command
claude mcp add --transport http assay https://app.assay.wiki/api/mcp/v2/mcpHTTP transport is the recommended path for remote MCP servers in Claude Code.
Cursor
- Open Cursor.
- Open Settings. You can use the gear icon, or open the Command Palette and search for MCP settings.
- Open Tools & Integrations or MCP, depending on your Cursor version.
- Click New MCP Server.
- Cursor opens mcp.json. Paste the JSON shown here, or merge the assay entry into your existing mcpServers object.
- Save mcp.json.
- Return to Cursor settings and click Refresh or Connect next to assay if Cursor shows that button.
- Complete OAuth in the browser if Cursor opens an authorization page.
- Open Agent mode and ask Cursor to list available Assay tools.
- Paste the Cursor prompt from the top of this page and run a read-only draft check.
mcp.json
{
"mcpServers": {
"assay": {
"url": "https://app.assay.wiki/api/mcp/v2/mcp"
}
}
}If Cursor rejects the direct URL field in your version, use the Gemini-style npx mcp-remote wrapper with the same Assay endpoint.
Microsoft Copilot
- Open Microsoft Copilot Studio.
- Open the agent that should use Assay.
- Click Tools.
- Click Add a tool.
- Click New tool.
- Choose Model Context Protocol.
- For Server name, type Assay.
- For Server description, type Approved GTM truth, brand guardrails, compliance checks, and receipts.
- For Server URL, paste https://app.assay.wiki/api/mcp/v2/mcp.
- For authentication, choose OAuth 2.0 if Copilot Studio asks.
- Choose Dynamic discovery if Copilot Studio offers it for OAuth.
- Click Create, then Next if the wizard shows another screen.
- Choose Create a new connection, finish the sign-in, and click Add to agent.
- Test the agent and ask it to use Assay for a read-only draft validation.
Server URL
https://app.assay.wiki/api/mcp/v2/mcpCopilot Studio supports Streamable HTTP MCP servers. Normal Microsoft 365 Copilot chat usually needs an admin-published agent or connector before users can invoke it.
If the Truth Graph is empty, seed facts additively from the app, approve them in review, then ask your assistant to ground or verify. Rich MCP hosts can render the seal inline; every host still receives the text receipt.
Questions
What your team will ask.
We already have RAG.
RAG retrieves text. It does not know what is approved, what is superseded, or who said so, and it leaves no receipt. Assay answers come from a governed graph with approval state and provenance, and the check leaves a verifiable mark.
Is this watching my team?
It watches the work, never the people. Capture is consent-gated. It offers, you decide, and nothing is recorded without a yes. Every capture is logged and keeps its author's name.
Another surface to govern?
The same governance you already set, traveling with the seat. No new rules to write. An editor's AI gets editor access, a viewer's AI gets viewer access, and every call lands in the same audit log.
Will it spam badges?
No seal on empty results, by contract. If no claims were checked, no checkmark appears.
Can the model just say it's verified?
It can say it. It can't sign it. The Seal is a signed receipt, and tampering with any part breaks it. The verify page shows what Assay actually signed.
Can a free workspace use it?
Yes. Free workspaces can use OAuth MCP read and verification tools with a metered cap. Scale and Enterprise unlock writes, drafts, approvals, and headless API keys.
Does this guarantee EU AI Act compliance?
No software can guarantee compliance by itself. Assay gives teams the evidence layer compliance reviews need: sourced claims, scoped access, human approval gates, guardrail results, receipts, and audit logs for AI-assisted GTM work.
Assay Everywhere
Connect your AI.
One connection, the same living truth in every tool. The web app stays the operating room for setup, review, and governance. Daily work stays where it already happens.