Protocol-Governed Knowledge Graph Memory Server — an MCP server that provides structured knowledge graph operations for LLM agents, implementing epistemic hierarchy principles: Evidence → Observation → Derivation, with full provenance tracking, confidence propagation, and version control via git checkpoints.
ENGRAM exhibits severe definition quality gaps across all 30 tools. While tool names follow verb_noun convention (add_observation, engram_checkpoint), descriptions are uniformly generic and fail to distinguish tools or guide LLM selection. 28 of 30 tools accept only 'payload_json' (a single string parameter), which violates the constrained-input pattern and forces LLMs to construct JSON without schema guidance. The one exception, engram_add_observation, has a detailed schema but its description ('Add an observation...') is only 60 chars and lacks context on when to use it vs. other observation tools. No tool documents output schemas, error recovery paths, or idempotency guarantees. Parameter descriptions are minimal (e.g., 'JSON payload containing...') and do not explain format, constraints, or dependencies. This design pattern (payload_json everywhere) is a red flag for agent usability, LLMs cannot validate JSON structure before calling, leading to runtime failures.
Add an axiom (foundational principle) to the knowledge graph
Add a conjecture (hypothesis) to the knowledge graph
Add a cornerstone (key concept) to the knowledge graph
Add a definition to the knowledge graph
Add a goal to the knowledge graph
Add a lesson to the knowledge graph
Add an observation to the knowledge graph with evidence provenance tracking
28 of 30 tools use single 'payload_json' parameter with no input schema validation. LLMs cannot construct valid JSON without seeing the expected structure, leading to runtime failures and retry loops.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 42 | 2026-07-28+ | v2 |
Add multiple observations to the knowledge graph in a single batch operation
Add a person node to the knowledge graph
Add a task to the knowledge graph
Advance to the next turn in the knowledge graph session
Ask a question in the knowledge graph
Create a checkpoint (version control point) in the knowledge graph
Record a contradiction between nodes in the knowledge graph
Derive new knowledge from existing nodes in the knowledge graph
Focus attention on specific nodes in the knowledge graph
Record tension between goals in the knowledge graph
Register an incident for a lesson in the knowledge graph
Create a link relationship in the knowledge graph
Deprecated backward-compat alias for engram_checkpoint
Mark a cornerstone as outgrown in the knowledge graph
Register an exemplar (example) for lessons and cornerstones
Remove an edge relationship from the knowledge graph
Report a feeling or emotional state to the knowledge graph
Resolve a question or uncertainty in the knowledge graph
Retract a statement or claim from the knowledge graph
Scan for emerging patterns in the knowledge graph
Mark a node as superseded by another in the knowledge graph
Remove focus from specific nodes in the knowledge graph
Update an existing task in the knowledge graph
All tool descriptions are generic and under 100 characters. None explain WHEN to use the tool, what distinguishes it from similar tools, or what the output contains. Descriptions like 'Add an axiom to the knowledge graph' do not guide LLM selection.
No tool documents its output schema or return type. LLMs cannot plan downstream tool calls or extract required fields (e.g., node_id for follow-up operations) without knowing what the response contains.
engram_nap is a deprecated backward-compat alias for engram_checkpoint. Exposing both creates confusion and wastes LLM reasoning cycles deciding between them. Remove the alias or clearly mark it as deprecated in the description.
No error handling guidance. Tools do not document what errors can occur, whether they are retryable, or what the LLM should do next. A failed tool call with no recovery path blocks the agent.