Valyd for AI Agents
Give your agent identity and verification as tools it can call — while the human stays in control. An agent can ask the signed-in user to approve a sensitive action with a face scan, read verification results, run web tasks on the user’s behalf, and request personal data that the person approves (or declines) in their own Valyd app. The agent gets capabilities; the user keeps consent.
Two different things share the word “agent” here — don’t confuse them:
Valyd MCP gives your agent verification and identity as tools it can call (a product
capability — it is not a documentation server). Agent-readable docs let any agent
read and learn these docs by plain fetch of llms.txt — no MCP, no server to
run. Use either, both, or neither.
Live today
A hosted MCP server (Streamable HTTP, OAuth 2.1 + PKCE) that turns verification and the Valyd web agent into callable tools for Claude Code, Claude Desktop, Cursor, VS Code, and any OAuth-capable MCP client. This is a product capability — not a docs server. No keys in config files: the user signs in with Valyd.
Valyd MCP — verification toolsPoint any agent at llms.txt to read and learn the whole docs by plain
fetch — the lightweight alternative to a “docs MCP”, with no server to run. Plus a full-corpus
llms-full.txt, per-page .md mirrors, OpenAPI 3.1 specs, and a Postman collection.
An agent (or your backend) requests raw user data; the human approves the exact fields in their Valyd app. Approved values come back sealed end-to-end to your key — Valyd stays blind to the plaintext.
Human-in-the-loop consentComing soon
These are on the roadmap and not yet available. Details will land here when they ship.
- Agent-initiated verification sessions — Coming soon
- Scoped per-agent credentials — Coming soon
- Token vault for agent integrations — Coming soon
Start here
One command in Claude Code; a two-line JSON block everywhere else.
Connect the MCP serverEvery tool — parameters, returns, and statuses.
MCP tool referenceOAuth 2.1 + PKCE, discovery via 401, and token requirements for code-first clients.
How MCP auth worksStart with the integration decision flow — the questions to ask the developer before writing code — then point your agent at llms.txt and call the APIs directly.
Agent integration guideQuestions? support@valyd.id