ContentsHide contents

MCP reference

Transport: Streamable HTTP at the server's mcp_endpoint. Authorisation per the MCP authorisation specification. Server name autonym.

Tool	Scope	Notes
`get_grant`	any	Read-only. Returns the grant's domains, levels, purpose and expiry.
`list_domains`	any read scope	Read-only. Domains this grant may read, and at what level.
`get_context`	`autonym:read:*`	Read-only. Counts as one read. Safety-critical entries are returned first.
`propose_entry`	`autonym:propose`	Non-destructive write. Lands in the review queue.
`confirm_entry`	`autonym:confirm`	Write. Uses elicitation where the client supports it.
`withdraw_proposal`	`autonym:propose`	Write. Removes an unconfirmed proposal.

`get_context`

json

{
  "name": "get_context",
  "description": "Read the person's context for the given domains at or below the granted disclosure level. Safety-critical entries are always returned first. Do not store beyond ttl_seconds.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "domains": { "type": "array", "items": { "type": "string" }, "minItems": 1 },
      "max_level": { "type": "string", "enum": ["L0", "L1", "L2", "L3"] },
      "keys": { "type": "array", "items": { "type": "string" } },
      "declared_only": { "type": "boolean", "default": false },
      "min_confidence": { "type": "number", "minimum": 0, "maximum": 1 }
    },
    "required": ["domains"]
  },
  "annotations": { "readOnlyHint": true, "openWorldHint": false }
}

Response

The same envelope is returned over REST.

Simplified and illustrative, not a schema-validated fixture. handle is shown here as the person's public handle for readability. The design issues each recipient a pairwise handle instead, so two recipients cannot compare what they hold and conclude they are reading the same person; the exact form is being settled against the reference implementation.

json

{
  "handle": "autonym:alex@autonym.me",
  "domain": "dining",
  "level": "L1",
  "issued": "2026-09-18T10:02:11Z",
  "ttl_seconds": 86400,
  "entries": [
    {
      "key": "allergy",
      "value": "tree nuts (severe)",
      "provenance": "declared",
      "safety_critical": true,
      "last_confirmed": "2026-09-01"
    },
    {
      "key": "diet",
      "value": "pescatarian",
      "provenance": "declared",
      "last_confirmed": "2026-08-30"
    },
    {
      "key": "drink_preference",
      "value": "flat white, oat milk, extra shot",
      "provenance": "declared"
    },
    {
      "key": "cuisine_affinity",
      "value": ["japanese", "levantine"],
      "provenance": "observed",
      "confidence": 0.78,
      "source": "delivery_history"
    }
  ]
}

Resource

autonym://context/{domain} is read-only, has the same ceiling as get_context, and counts as a read.

Prompt

autonym_briefing returns a one-paragraph summary of agent at L1 and communication at L1 for system-prompt use.

Errors

Tool errors return isError: true with a Problem Details object in the content.