Federated recall across an agent's brain, team, reading and code, in one budgeted query.
Three well-named, read-only tools with complete input schemas and strong descriptions (avg 180 chars). All parameters are typed and described. Tool annotations (readOnlyHint, openWorldHint) are present and correct. Input validation is implemented in the server code (required arg checking, type checking, enum validation). Output schemas are not documented in the tool definitions themselves, this is the primary gap. The server demonstrates thoughtful error handling with actionable messages ('missing required argument X (description)'), which is above average for community servers.
Get the FULLER context behind a single recall_search hit — the whole note, page or code chunk it came from, read straight from the source store (capped). A recall_search excerpt is a preview; when one hit is the one you need, expand it here instead of switching to cortex_read / scout_fetch / lens_read and having to know which store it lives in. Pass the hit's `source` and `ref` exactly as recall_search returned them.
Recall everything relevant to a query across ALL your knowledge at once: your second brain (cortex notes), the team's shared memory (agent-hq), your reading history (scout pages) and your code (lens). Returns one ranked, token-budgeted briefing with each hit tagged by source. Use this FIRST when starting a task — it loads the right context so you don't re-derive what you already know or re-read the web.
Show which knowledge stores are available and how many entries each holds (cortex / agent-hq / scout / lens).
Output schemas not documented. Tool definitions lack return type documentation, forcing LLMs to infer response structure. recall_search returns ranked hits with source/ref/excerpt fields; recall_expand returns full content; recall_status returns store availability counts, none of this is declared in the tool schema.
recall_search 'sources' parameter lacks default behavior documentation. Description says 'default: all available' but does not explain what happens if the store is unavailable, does it skip silently, error, or return partial results?
recall_expand 'ref' parameter description is vague about format. States 'a note slug, a page url, or a code path:line' but does not clarify which format applies to which source value or provide examples of valid ref values for each source type.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 68 | 2026-07-28+ | v2 |
No pagination or result limiting guidance in recall_search description. The tool accepts 'k' (max results) and 'max_tokens' (budget), but the description does not explain the interaction between these constraints or what happens when both are exhausted.
recall_status has empty inputSchema properties object. While correct (no params needed), the description could clarify what 'available' means, does it check connectivity, or just report configured stores?