Skip to main content
CID222 Docs

MCP servers

What the AI Inventory MCP Servers page does in 0.4 — a placeholder — and how CID222 does and does not see Model Context Protocol traffic today.

  • Version: 0.4
  • Role: admin_user

AI Inventory → MCP Servers is reserved for managing the Model Context Protocol (MCP) servers your organisation runs. In 0.4 it is a placeholder: the page renders a Coming Soon tile and a link to the protocol's own site, and nothing else.

What the page shows

The page has a header — MCP Servers, with the subtitle Model Context Protocol server management — and one centred tile reading Coming Soon and MCP server management will be available in a future update. The tile's only control is Learn More, which opens https://modelcontextprotocol.io in a new tab.

There is no table, no form, no dialog and no filter. The page issues no requests to the gateway, so nothing on it can succeed or fail, and it writes no event to Event logs. It is offered to superadmin, admin_user and viewer, and it is not licence-gated — every tier shows it.

Warning

The in-product help drawer describes this page as "Register Model-Context-Protocol servers so agentic tools can be brokered — and inspected — through the gateway (Professional and Enterprise packages)." None of that is true in 0.4: nothing can be registered, nothing is brokered or inspected, and the page carries no licence key.

How CID222 sees agentic traffic today

CID222 inspects the model calls an agent makes, not the tool calls it makes. When an MCP-speaking client reaches a model through the gateway, the inline proxy, the browser extension or the Claude Code plugin, that prompt and response go through the normal detection pipeline and land in All Detections like any other traffic. The MCP conversation between the client and its own MCP servers does not pass through CID222 at all, and no component in the product parses it.

That leaves one honest gap: an MCP server can hand an agent data that never crosses a CID222 inspection point, and CID222 has no record of it.

This page is not the CID MCP server

Two unrelated things in CID222 are called an MCP server.

AI Inventory → MCP Serverscid-mcp-server
What it isThe placeholder page described aboveA read-only MCP server over CID222's own Postgres
Who uses itNobody yetThe LLM Risk Analyst, and only it — the analyst holds no database connection string
Where it runsIn the dashboardA container on 127.0.0.1:8020, opt-in through the llm-analyst compose profile
Licence keyNonemcp.server, in the Professional and Enterprise tiers

cid-mcp-server exists so the analyst can read detection evidence through a database role that holds SELECT on a fixed list of tables and columns, rather than through a general connection. It has no dashboard surface, it is never exposed to a customer's own agents, and the mcp.server licence key does not unlock the MCP Servers page. Buying the key changes nothing about what this page shows.

Limits and known gaps

  • The page manages nothing. There is no MCP server record anywhere in CID222 — no table, no entity, no API. A "coming soon" here means not started, not partially built.
  • Governance cannot hold one either. The AI asset inventory declares mcp_server as one of its five asset kinds, but discovery never produces that kind and there is no endpoint that creates an asset by hand, so the vocabulary has no producer. The governance detail drawer says so directly, reporting the kind as unsupported with a reason rather than an error.
  • MCP tool traffic is not inspected. Nothing in the detection pipeline reads the protocol. Any claim that CID222 brokers MCP is a claim about a future release.
  • The page registry names a controller this page never calls. The mcp-servers row cites models.controller.ts, copied from its sibling Models row; the page makes no request at all. Treat the roles it grants as inherited defaults, not as evidence of a backend.

Note

If you need an inventory of the MCP servers in your estate today, record them outside CID222. Nothing in the product will discover, import or reconcile them.

  • Models — the half of the AI inventory that is implemented: the models each connected provider offers, and what the gateway routes to.
  • Governance inventory — where discovered AI assets are owned and classified, and where an MCP server would eventually belong.
  • All detections — what CID222 does record about agent traffic that passes through an inspection point.

Last updated on

On this page

Download PDF