Read evidence without executing writes
Use your configured MCP endpoint and credential. All five tools are read-only with respect to upstream APIs. The authenticated identity determines tenant scope; there is no tenant argument. Discover tools/list and the signed inputSchema/outputSchema before calling. Tool arguments reject unknown fields. Managed MCP access is not available.
Tool inputs and outputs
Every successful tools/call result includes content text containing JSON and structuredContent: {"result": value}. null or empty evidence is not proof of safety.
| Tool | Arguments | result value |
|---|---|---|
| lookup_contract | {tool: string (1–256 characters)} | Selected {tool, version, schema, renames?}, or null. renames are candidate mappings with confidence and identityConfirmed. |
| query_drift | {tool: string, limit?: integer 1–100 (default 20)} | Returns at most 100 versioned drift events; [] means no returned evidence. |
| fetch_audit | {id: string (1–256 characters)} | One versioned audit artifact or null; evidence is scoped to the authenticated tenant. |
| simulate_repair | {tool: string, args: object, mode?: observe | repair | strict (default repair)} | {applied, dispatched: false, changes, args, reason}. applied means a simulated argument change only; never a vendor write. Observe leaves args unchanged. |
| coverage_report | {tools: string[] (1–100 entries, each 1–256 characters)} | {status: "unvalidated", reason, tools: [{tool, contract, observations}]}. Observations do not establish production safety. |
Call example
This is the JSON-RPC body for a configured MCP transport. Supply application/json, Accept: application/json, text/event-stream, MCP-Protocol-Version and your authorized credential. The read-only local interface guide includes an executable initialization/call sequence.
{
"jsonrpc": "2.0",
"id": 2,
"method": "tools/call",
"params": {
"name": "coverage_report",
"arguments": {
"tools": [
"vendor.read"
]
}
}
}Example structured result
Illustrative successful coverage result. The text content represents the same result value; tool output schemas are available in tools/list and the independently verified signed description.
{
"content": [
{
"type": "text",
"text": "{\"status\":\"unvalidated\",\"reason\":\"Observed coverage is not a production drift-safety guarantee.\",\"tools\":[{\"tool\":\"vendor.read\",\"contract\":false,\"observations\":0}]}"
}
],
"structuredContent": {
"result": {
"status": "unvalidated",
"reason": "Observed coverage is not a production drift-safety guarantee.",
"tools": [
{
"tool": "vendor.read",
"contract": false,
"observations": 0
}
]
}
}
}Three failure layers
HTTP 401/403/413/503 describes authentication, authorization, request-size or service-availability failure; see the transport reference. A JSON-RPC error describes an invalid method or parameters. A tool result with isError: true and text "Tenant evidence unavailable or invalid." describes an evidence failure even if HTTP is 200. Do not interpret missing structuredContent or an isError result as no drift. Fix invalid input, restore authorization, or retry evidence reads with a capped delay; none of these responses authorizes a vendor write.