mcp-server-fleet

We need screenshot-ready views of org-wide MCP server usage and relationships, backed by deterministic synthetic datasets so layouts and captions can be tuned without relying on live telemetry.

The seed dataset for this cycle is a 6-row MCP server table screengrab; this cycle recreates that dataset as a QMap-ready synthetic scenario and renders three QMap views.

Deliver one runnable experiment under dev/experiments/c27.mcp-server-fleet/ that can render:

1. Flat Fleet — MCP servers only, uniform sizing.
2. Usage-Sized Fleet — MCP servers only, size driven by call volume proxy (sessionsCount).
3. Callers View — MCP servers with the agents and users that called them.

All three views must be deterministic (stable positions + stable call graph given the scenario seed).

• Default scenario reproduces the 6 server rows exactly:
  • Filesystem (1.0.0, stdio, tools 11, skills 0, sessions 5)
  • web-search (0.4.2, sse, tools 3, skills 12, sessions 9)
  • github (1.2.0, stdio, tools 8, skills 10, sessions 100)
  • postgres (0.9.1, stdio, tools 5, skills 4, sessions 82)
  • slack (0.3.0, http, tools 11, skills 78, sessions 2)
  • internal-docs (0.1.0, sse, tools 1, skills 2, sessions 18)
• Flat fleet view renders servers only and shows per-server metadata in an overlay on selection.
• Usage-sized view visibly scales github/postgres larger than others.
• Callers view shows both users and agents connected to servers, grouped by entityType.