Knowledge-lifecycle MCP: distill agentsync + context-keeper events into memory, recall across projects, promote local -> team -> org
cambium has 10 tools with complete input schemas and descriptions, but quality is uneven. Naming is verb-forward and clear (setup, distill, recall, promote, endorse, deprecate, list_items, get_item, create_item, update_item). Descriptions are substantive (100 - 250 chars), explaining WHAT and WHEN. However, parameter descriptions are sparse or missing context. Output schemas are not documented, LLMs cannot infer what fields to expect from recall(), list_items(), or get_item(). Error handling is absent: no guidance on retryability, user-fixable vs fatal errors, or recovery paths. The tool set is well-composed (each does one thing), but lacks the polish of production-grade tools.
Manually create a knowledge item in local scope with content, optional tags, and metadata. Assigns a unique ID and records creation timestamp.
Mark a knowledge item as deprecated with an optional reason. Deprecated items are no longer recalled but remain in the store for provenance and reversibility.
Turn agentsync coordination events and context-keeper project memory into cambium knowledge items. Reads agentsync branch and .context/ files, creates/updates items in local scope with source provenance.
Vouch for a knowledge item, increasing its trust score and enabling promotion to org scope. Endorsements are recorded with the endorser's agent_id and timestamp.
Retrieve a single knowledge item by ID with full metadata, source provenance, trust history, and recall log.
List all knowledge items in a given scope with their metadata (id, content, scope, status, trust scores, source provenance).
Output schemas not documented. LLMs cannot infer what fields recall(), list_items(), get_item(), and promote() return. This forces agents to guess at response structure and risks failed downstream tool chaining.
No error handling guidance. Tools lack recovery hints (e.g., 'item not found, try recall() to search' or 'promotion failed due to insufficient recalls, endorse the item first'). Agents cannot self-correct on failures.
Parameter descriptions lack context. 'scope' enum is documented but not explained when to use 'local' vs 'team' vs 'org'. 'limit' has no bounds (min/max). 'endorsement' is marked required for team→org but this dependency is not stated in promote() description.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2026-07-28+ | v2 |
Graduate knowledge from local -> team -> org as it earns trust (recalls >= CAMBIUM_PROMOTE_RECALLS for local->team, endorsement required for team->org). Pushes to team branch or org repo with provenance and reversibility.
Search cambium knowledge across scopes (local, team, org) by lexical relevance. Returns items ranked by token overlap with the query, filtered by RELEVANCE_FLOOR (0.2). Scoped to the caller's project and promotion tier.
Configure cambium: set CAMBIUM_REPO, CAMBIUM_AGENT_ID, and optional org/team settings. Returns structured guidance on what's set, what's missing, and the cost of each gap.
Update an existing knowledge item's content, why, example, or tags. Preserves ID and source provenance, updates modified timestamp.
No pagination documented for list_items(). If a team has hundreds of knowledge items, returning all of them blows context. Should accept offset/limit and return total_count or next_cursor.
Dry-run or confirmation missing for irreversible operations. promote() and deprecate() modify state permanently. No confirm_before_execute pattern or dry-run mode to let agents preview changes before committing.