Agents, assistants and MCP servers

Agents act, not just answer. What they can do depends on the tools they can call, so the inventory records those tools.

Register an agent

1. 1
   Go to AI systems → Agents & assistants and register a new agent, or attach a profile to a system you already have.
2. 2
   Say what it is and how independently it acts: a person approves each step, it acts within limits, or it acts on its own.
3. 3
   Add the MCP servers it connects to. Search 55 common servers (GitHub, Slack, Stripe, Postgres, Salesforce…) and pick them; V77 Trust fills in the tools each one usually exposes and the data it reaches. Remove any tools your setup doesn't give the agent. For your own servers, type the name and list its tools.
4. 4
   Record the controls: how a person monitors or stops it, whether there's a kill switch, whether every action is logged, and any limits (for example "refunds up to €200").

Checks

Each tool is labelled by what it can do: moves money, runs code, sends messages, changes data or reads only. V77 Trust then checks, for example:

• agents that can move money or run code without a person approving each step;
• missing kill switch or action logging;
• high-risk systems with unlimited autonomy;
• custom MCP servers whose tools aren't listed;
• whether the agent has been red-team tested for prompt injection.