Proof-of-Intent for AI agents
A remote always-on MCP Guard, HTTP Action API and local wrapper that let agents ask SecureStamp before acting. It authorizes the next step; it never moves money, deletes data or executes the operation.
Authorizer, not executor
Tenant-scoped API keys and clients
Signed Action Receipts
POST https://mcp.securestamp.online/mcp
Authorization: Bearer ss_live_...
{ "jsonrpc":"2.0", "id":1, "method":"tools/list" }authorize_action
Returns an Action Verdict, Safe Next Step, and receipt before a sensitive action.
analyze_message_intent
Maps abstract signals to sensitive action intents. Signals only; no raw message text.
verify_counterparty
Checks whether a sender or counterparty matches the tenant registry.
create_action_challenge
Creates a manual dual-control challenge for a known counterparty.
get_safe_next_step
Computes the safest next step from declared registry facts.
issue_action_receipt
Issues a signed Action Receipt for already-computed facts.
Parable 3
The agent pauses before changing an account
A workflow receives a vendor bank-change request. Instead of updating the account, it calls authorize_action through the remote MCP Guard or HTTP Action API with a hash and abstract signals. SecureStamp returns needs_confirmation and a Safe Next Step: challenge the counterparty.
The agent detects a sensitive action
It identifies a bank-account change and sends only structured signals plus a sourceMessageHash.
SecureStamp checks tenant facts
The API key scopes the lookup to one owner graph, declared policies, and registered instructions.
The agent receives a verdict
The response is a decision record with Safe Next Step and receiptId, not permission to execute.
Humans or counterparties confirm
If needed, create_action_challenge opens a manual confirmation path before the operation proceeds.