The agent awareness layer - an MCP server that provides awareness surfaces for AI agents through channels, events, states, tasks, and notes
Agentenna exposes 4 read-only tools with clear verb-noun naming (list, listen, rewind, inspect). All tools have descriptions and input schemas with typed parameters. However, parameter descriptions are sparse or missing for optional fields, output schemas are not documented, and error handling guidance is absent. The tools are well-composed for a single domain (channel awareness), but lack the depth of documentation and error recovery patterns expected in production tools.
Return full event detail, including payload when retained.
List Agentenna channels with summary counts.
Return the channel's composed awareness surface as compact XML. One read: matching instructions/learnings, current states, open tasks, and recent events in the <agentenna-awareness> wrapper.
Search historical channel events and return compact XML.
Output schemas not documented. Tools return XML strings or dicts, but LLMs cannot infer field structure. agentenna_listen and agentenna_rewind return .render() output (XML) with no schema; agentenna_list_channels and agentenna_inspect return dicts with undocumented fields.
Optional parameters lack descriptions. agentenna_rewind has 7 optional params (query, window, severity_min, kind, entity, source) with no descriptions explaining what each filters or how to use them. LLMs cannot reason about when to pass these.
No error handling or recovery guidance. Tools do not document what happens on invalid channel names, missing event IDs, or malformed time windows. No guidance for LLMs on retryability or next steps.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | 2026-07-28+ | v2 |
Parameter constraints not documented. 'window' accepts time strings (e.g. '30m') and 'limit' is unbounded. No description of valid window formats, min/max for limit, or what happens if invalid values are passed.
No pagination or result-limiting guidance. agentenna_rewind and agentenna_listen accept 'limit' but do not document max results, whether pagination is supported, or how to handle large result sets.