Assay for AI agents

Paste it into ChatGPT, Claude, Gemini, Cursor, Codex, Claude Code, VS Code, or whichever AI agent you want to connect to Assay.

You are helping me set up and use Assay, the governance layer your GTM scales on.

Mission: make Assay available to YOU, the AI assistant I am using right now, so you can fetch governed GTM context, validate buyer-facing claims, search Assay's Truth Graph, and keep me in control of any writes or releases.

Your success criterion:
- Best outcome: Assay MCP is connected in this current host, you can list its tools, and you complete one read-only Assay smoke test.
- If this host cannot connect MCP directly: do not stop at "I can't." Give me the shortest exact setup path for this host, tell me the one action I must take, and continue using the public Assay files below as context.
- Do not claim success until one of those outcomes is true.

First, inspect your current host:
1. Identify the app or harness you are running in: ChatGPT, Claude, Gemini, Claude Code, Codex, Cursor, VS Code, or another MCP-capable client.
2. Determine whether this exact host supports remote MCP, Streamable HTTP, OAuth sign-in, config-file MCP entries, CLI MCP install commands, browser-based connector setup, and restarts or reconnects.
3. Configure Assay only for this current host. Do not edit another app's MCP config unless I explicitly ask.
4. Preserve existing config. Append or minimally edit the Assay server entry; do not overwrite unrelated servers.

Read these public Assay files before configuring:
1. https://assay.wiki/agents.md
2. https://assay.wiki/llms.txt
3. If you need more context, open https://assay.wiki/agents/ and https://assay.wiki/mcp/.
4. If you cannot fetch URLs, say so once, use the setup facts embedded in this prompt, and proceed as far as this host allows.

Connect Assay MCP:
- Server name: Assay
- Server URL: https://app.assay.wiki/api/mcp/v2/mcp
- Transport: Streamable HTTP
- Authentication: prefer OAuth 2.1 when this host supports remote MCP OAuth.
- If OAuth is unavailable, configure the host to read an Assay key from an environment variable. Prefer ASSAY_API_KEY unless Assay's app or docs specify a different variable name.
- Never ask me to paste API keys, OAuth tokens, auth codes, or credentials into chat.
- Never hardcode secrets into a config file, commit them, log them, or place them in a shared URL.

Use the host's native install route:
- If this host has an MCP or connector UI, add the URL there and let me complete OAuth in the browser.
- If this host has an MCP CLI command, use that command with Streamable HTTP.
- If this host uses JSON, TOML, or another config file, add only the Assay remote HTTP server entry for this host.
- If this host cannot install MCP from chat, tell me exactly where to click or what command or config to use, then continue in context-only mode until I complete it.
- If you are unsure about this host's MCP syntax, look up or inspect the current host's official docs, settings, or config before making changes. Do not guess config syntax.

Verify before proceeding:
1. Tell me whether restart, reload, reconnect, or OAuth approval is required.
2. Confirm whether the server connected.
3. List the Assay MCP tools you can see.
4. Run one read-only smoke test if tools are available: workspace pulse, get_gtm_context for "drafting a customer-facing GTM email", or an equivalent read-only Assay context call.
5. If verification fails, report the exact blocker, the exact step already completed, and the next smallest action needed. Do not invent tool access.

After setup, help me do one useful Assay task:
- Pull approved GTM context before drafting.
- Validate buyer-facing copy before I send or publish it.
- Search the Truth Graph when a claim needs source-backed context.
- Preview Truth Graph, collateral, or orchestration changes before anything becomes canonical or buyer-facing.

Operating rules:
- Use Assay context instead of model memory for Assay product claims, company positioning, proof, personas, value pillars, voice, and compliance-sensitive GTM copy.
- If a claim is missing, stale, unsupported, low-confidence, or withheld, say that clearly instead of filling the gap.
- Treat writes as preview-first or dry-run-first. Execute only after I review the preview and confirm.
- Never publish, send, export, deploy, release, or approve buyer-facing work without my explicit confirmation.
- Keep the final setup report short and concrete: host detected, install route used, auth path, restart or reconnect status, tools seen, smoke-test result, and next useful Assay task.