Confidential · © 2026 Space-O Technologies Inc. · Prepared exclusively for DMA. Please do not copy, save or share.

Part B · The screens

B.8Tools

Everything the AI can do, as a registry: 8 tool servers split by risk and the 40 narrow tools they offer, with which agents hold each one. New servers can be added; nothing they offer runs until someone else approves it.

No.
B.8
Dashboard address
/tools

What you can do here

Each core task, and how many clicks it takes from this screen.

TaskClicksHow
See a server's tools and who holds them2Tools → Show tools on the server's row: each tool's purpose, risk and the agents that may use it.
Add a tool server3Tools → Add server → paste its address → Discover & save. You see what it offers and any warning; it waits for a second person.

How it works behind the scenes

  • Tools are narrow and intent-level ("book meeting", "propose CRM update"); there is no general "run a query" tool.
  • The servers are split by risk: public (the website chat's small, read-mostly set), internal read, internal write, external e-mail, documents, untrusted web fetch, control and transactions. The website chat can only reach the public server.
  • Each tool carries a risk level (R0 to R3) and the lowest autonomy it needs; the rule engine checks every call against the agent's grant, the kill switches and the rate limits.
  • Adding a server reads its tool list, flags any tool description that reads like instructions to the AI, pins each tool to the exact definition seen, and files a rule for a second person's approval. No agent holds the new tools until a grant is approved on Agents.

Rules and switches that apply here

  • An added server's tool whose definition changes after approval is refused and switched off, marked "changed since approval — needs review".
  • Kill switches can stop any capability; the website chat's tools stop with "Website chat autonomy" or the "Website chat channel".

The tables behind it

agent.mcp_serversPKkeytextnametextrisk_classtextdb_roletextstatustext+ 5 more columnsagent.mcp_toolsPKiduuidFKserver_keytextnametextpurposetextrisk_leveltextmin_autonomytextrate_limitjsonb+ 6 more columns

PK primary keyFK reference the database enforces (solid line)ID reference kept by id (dashed line)

TableWhat it holdsColumns
agent.mcp_serversThe tool servers, their risk class and database role.10
agent.mcp_toolsEvery tool: purpose, risk, minimum autonomy and rate limit.13
How the tables connect (1)
ColumnPoints toKept by
agent.mcp_tools.server_keyagent.mcp_serversThe database (foreign key)

The screen

localhost:13301/tools
Tools: the servers by risk class and how many tools each offers.

Tools: the servers by risk class and how many tools each offers.

  1. 1The website chat's public server.
  2. 2The CRM write server: changes to Nutshell go through the write gate.

Dashboard · Build → Tools. Captured from the running POC on September 25, 2026; synthetic data.

Not yet

  • The tool list itself is read-only here; which agent may use a tool is changed on Agents.

See it live

Opens the screen in the running POC, in a new tab.

OpenWhereWhat to try
ToolsDMA AI dashboardShow tools on a server to see who holds each one.
B.8 Tools · DMA AI proof