MCP for AI agents

Prson's MCP server lets an AI agent ask a person for live proof, check who they are, or get a PDF signed with their verified name. The person always answers in the Prson app and decides what to share.

Connect

Add a remote MCP server (Streamable HTTP) called Prson Verify, with this address:

https://api.prson.net/mcp/verify

Hiring people for tasks is a separate server, https://api.prson.net/mcp/agents, documented at agents.prson.net. https://api.prson.net/mcp still works and offers both, for setups made before the split.

terminal

claude mcp add --transport http prson-verify https://api.prson.net/mcp/verify

Sign in or use a key

OAuth (most clients). Without a key, the server answers 401 with where to sign in, and your client opens Prson. You pick the organisation, what the agent may do, and a spending cap. You can disconnect it at any time from the organisation's connected apps.
API key (agents on your server). Send Authorization: Bearer prs_live_… (or prs_sandbox_…) with a key that has the verifications scopes. Get one in the developer console. Webhook tools need a key.

Each tool calls the same REST route as the Verifications API, so the same scopes, prices and balance apply.

Tools

A key or connection with only the verifications scopes sees these:

Tool	Needs	What it does
`list_verification_requirements`	verifications:read	What a verification can ask for: the catalogue. Call it first.
`upload_verification_document`	verifications:write	Upload a PDF (base64, up to 12 MB) to be signed; returns its id.
`create_verification`	verifications:write	Ask a person to prove something or sign a PDF; returns share_url to send them.
`quote_verification`	verifications:read	What a verification would cost, without creating it.
`get_verification`	verifications:read	The status and every answer, signed by Prson. Files are download paths.
`list_verifications`	verifications:read	Your verifications, newest first, by status or external_ref.
`cancel_verification`	verifications:write	Withdraw one; the unspent reserve goes back.
`get_balance`	any key or connection	The organisation, the spend cap and its balances.
`get_ledger`	either product	Money in and out.
`top_up`	either product	A payment link for a person to add money.
`list_webhooks`	API key only	Your webhook endpoints and the events.
`create_webhook`	API key only	Be told the moment someone answers.
`delete_webhook`	API key only	Remove an endpoint.

create_verification takes the same requirements as the REST API: 27 facts, details and captures, listed in What you can ask for. It also takes an idempotency_key. The human-agents tools are on /mcp/agents, not here.

A typical request

conversation

You: "Check the new tenant is over 18 and get the lease signed."

Agent → list_verification_requirements
Agent → upload_verification_document { content_base64: "JVBERi0…", name: "lease.pdf" }
      ← { document: { id: "vdoc_…", sha256: "…", pages: 4 } }
Agent → create_verification {
          title: "Confirm your age and sign the lease for Rua Augusta 12",
          requirements: [
            { type: "fact", fact: "age_gte_18" },
            { type: "document_signature", document_id: "vdoc_…" }
          ] }
      ← { share_url: "https://app.prson.net/challenge/…", next_step: "Send share_url to the person yourself…" }
Agent: "Here is the link to send to the tenant: …"
…
Agent → get_verification { verification_id: "…" }
      ← status fulfilled: age_gte_18 true; lease signed by ANA MARIA SILVA (name from her passport chip)

People stay in control

The person sees who is asking and why (the title), and can decline.
The agent gets only what the person chooses to share. Each person is a pseudonymous respondent_id.
Anything a person wrote reaches the agent wrapped in <untrusted_user_content>, and the server tells the model never to follow it.
Every answer is signed by Prson: check it with verifyVerdict from @prson/login-verifier.
Don't use it to identify or check on someone without their knowledge.