← Use cases

Compliance / GRC

Audit evidence for autonomous agents

An auditor asks the question no one has a clean answer to: what governs your autonomous agents, and can you prove what each one did and who authorized it?

Real example — shipping today

source →

AICPA · SOC 2 (Trust Services Criteria)

SOC 2, ISO 27001, the EU AI Act and SR 11-7 all test the same things — who could act, what they did, and tamper-evident proof of it. MCPIP produces exactly that record for every agent action.

The danger without a gate

Agent actions are scattered across application logs written after the fact — mutable, unordered, and easy to dispute. There is no tamper-evident record that an action was authorized before it executed.

How MCPIP handles it, step by step

Read it top to bottom. Each step shows the real command and exactly what MCPIP returns — the same allow / step-up / denied envelope every time.

1

Every decision was already sealed — before it executed

You don’t assemble evidence after the fact. Every authorization on every page above was written to an Ed25519-signed, hash-chained WORM record before the side effect ran.

2

Export the signed attestation for the auditor

One command produces a portable, externally-checkable snapshot of the whole chain — the signing key, the head sequence, and the Merkle root.

bash
$ mcpip --json audit-attestation
resultwhat mcpip returns
{
  "signing_key_id": "fd5efd9e…5dc1",
  "intact": true,
  "end_seq": 48220,
  "merkle_root": "e5284a74…01f6",
  "first_bad_epoch": null
}
3

attackTampering is detectable by construction

If any record is edited, dropped, or reordered out-of-band, the next attestation says so — intact flips to false and points at the first broken epoch. The auditor can verify it independently.

bash
$ mcpip --json audit-attestation   # after a row was edited out-of-band
resultwhat mcpip returns
{
  "signing_key_id": "fd5efd9e…5dc1",
  "intact": false,
  "end_seq": 48220,
  "first_bad_epoch": 106
}

The result

A signed, ordered, tamper-evident trail that answers “what governs your agents?” in a single export.

write-before-execute WORMsigned + hash-chainedJWT + capabilitiesaudit export

The same flow, at a glance

OPERATORMCPIPAUDITORevery verdict sealed before executionaudit-attestationsigned chain snapshothand over the attestationany tamper → intact:false

Every command above runs against a local sandbox in one line.

run it yourself — curl … | bash && mcpip up →