The governance is the product, and it is specific: a write runs dry by default and mints a single-use, expiring plan token that has to be confirmed before anything changes; pricing, policy, proof and identity records are refused for autonomous mutation altogether, and unmapped kinds fail closed. Check which entitlement lane your plan is in before wiring an agent that needs to write — read and verification tools are the Free tier surface.
A remote MCP server for go-to-market teams. Positioning, proof points, brand voice, personas and validation rules live in a governed Truth Graph, and the server hands that context to MCP clients instead of copying company facts into each agent.
- Truth Graph search with provenance, so a fact arrives with where it came from
- Draft content validated against approved facts and guardrails before it ships
- Governed mutation: create and update records through dry-run and human-approved workflows
- Grounded collateral generated into quarantine, with preview and export gates, and re-rendered when the underlying story changes
- An AI search and GEO workspace — briefs, artifacts, evidence, goals, hosted sites and approval-gated publishing
- Readiness simulations and pre-call briefs
- 48 registered tools in total, with `tools/list` on the live endpoint as the current surface
An Assay account. The endpoint is `https://app.assay.wiki/api/mcp/v2/mcp` over streamable HTTP — Claude Code adds it with `claude mcp add --transport http assay-truth-graph`, and any client takes the same URL. Interactive clients get a browser OAuth prompt on first use; headless agents authenticate with an org API key from Assay Settings. Access is tiered: `mcp_access` covers read and verification tools from Free upward, `mcp_control_access` covers write and run tools from Scale upward.
One command plus a key — claude mcp add --transport http assay-truth-graph https://app.assay.wiki/api/mcp/v2/mcp, then supply credentials