Open-source mission control for AI agent fleets — chat to provision, agents to operate, humans to approve. Knowledge graph, governance, swarm coordination, MCP-native runtime, and the underlying fleet substrate.
This is a STDIO-only example server with 8 basic tools. All tools have descriptions (10-50 chars, below the 50-200 char LLM-optimized baseline) and input schemas with typed parameters. However, descriptions are minimal and lack context for LLM selection. No output schemas documented. No error handling guidance. No tool annotations (readOnlyHint/destructiveHint). The write_file tool is marked WRITE but has no confirmation or dry-run pattern. Tool names are clear (add, subtract, multiply, divide, read_file, write_file, list_directory, get_weather) but descriptions do not explain WHEN to use each tool or what happens on error. Parameters lack format constraints (e.g., get_weather 'location' accepts any string; no validation hints). No pagination for list_directory despite potential for large results.
Adds two numbers together
Divides two numbers
Gets current weather information for a location
Lists the contents of a directory in the allowed paths
Multiplies two numbers
Reads the contents of a file from the allowed paths
Subtracts two numbers
Writes content to a file in the allowed paths
Descriptions are too brief (10-50 chars vs. 50-200 char baseline). LLMs cannot determine WHEN to select each tool or what happens on error. E.g., 'Adds two numbers together' lacks context for multi-step math planning.
No output schemas documented. LLMs cannot plan downstream tool calls or extract structured data. E.g., does get_weather return {temp, humidity, condition} or a free-text string?
write_file is destructive (WRITE risk) but has no confirmation, dry-run, or error recovery guidance. Agents may overwrite files without warning.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Agents cannot infer which tools are safe to retry or have side effects.
list_directory has no pagination (limit, offset, next_cursor) or result count cap. Large directories could blow context window.