delegated-authority

Delegated Authority Skill

MCP Endpoint

Field	Value
Server	Tenzro L1 core MCP server
Testnet endpoint	https://mcp.tenzro.xyz/mcp
Local endpoint	http://localhost:3001/mcp
Tool	authorize_tool_call
Underlying RPC	tenzro_authorizeToolCall

What this is for

Before running a tool that changes something — an edit, a shell command, a transfer, a contract call — ask whether you are allowed to.

The answer does not come from local configuration. It comes from the delegation grant your controller signed: which parties you may act as, which templates and commands you may use, what you may spend per call, and which actions their oversight claim requires a human to approve.

This matters because a policy held in a config file cannot constrain the process running you — anything able to edit that file can remove the gate. A grant is signed, lives on the network, and answers the same way whether you are running inside a coding harness or autonomously.

When to call it

Call it when the tool would change state or spend:

• writing or deleting files outside a scratch directory
• shell commands with side effects
• any transfer, contract call, or paid resource invocation
• anything acting on behalf of a party you were delegated

Do not call it before reads, searches, or listing operations. A gate on every read makes the check noise and trains you to ignore it.

The three answers

authorize_tool_call(tool="edit_file")
→ {"decision": "permit", "tool": "edit_file", "resource_id": "tool:edit_file", "wallet_id": "..."}

permit — proceed.

require_approval — a human must approve first. This is an outcome, not a failure. Do not retry, do not route around it, and do not report it as an error. Surface the approval_id to the human, stop that line of work, and continue with something else if there is something else to do.

→ {"decision": "require_approval", "approval_id": "apr_...", "reason": "..."}