Biologically-inspired memory for AI agents. Exposes hippo memory as MCP tools over stdio transport. Supports decay, retrieval strengthening, consolidation, hybrid search, and session continuity.
Hippo Memory has 23 well-named tools with consistent verb_noun patterns (hippo_recall, hippo_remember, hippo_forget). All tools have descriptions (avg ~150 chars, within baseline 34-392 range). Input schemas are present with typed properties. However, many parameter descriptions are terse (10-30 chars), falling below the 72-char baseline for param annotations. Output schemas are not documented, LLMs cannot predict response structure. Error handling is absent; no recovery guidance or categorization. Security concerns: no visible permission gates on destructive ops (hippo_forget, hippo_consolidate). Composition is strong, tools chain well (recall → remember → outcome). Overall: solid naming and basic schemas, but weak descriptions, missing output docs, and no error guidance.
Assemble a comprehensive context from memories matching a query, including related memories and continuity information.
Automatically extract and store lessons from git history or other sources.
Run memory consolidation to merge related memories, remove duplicates, and optimize the store.
Retrieve detailed information about a specific memory or explore related memories in depth.
Estimate the token count for a given text using the configured tokenizer.
Delete a memory from the store. Irreversible operation.
Retrieve the current hippo configuration.
Output schemas not documented. LLMs cannot predict response structure (fields, types, pagination). Baseline: 100% of A+ tools document return types.
Parameter descriptions are terse (10-30 chars). Baseline: 72 chars avg. Examples: 'ID of the memory to delete' (28 chars), 'Search query' (12 chars). Expand with format, constraints, and when to use.
No error handling or recovery guidance. Destructive tools (hippo_forget, hippo_consolidate) lack confirmation or dry-run support. No actionable error messages documented.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
Retrieve a specific memory by ID with full details.
Retrieve statistics about the memory store (total memories, average strength, etc.).
List all unresolved memory conflicts.
List all memories in the store with optional filtering by tags, scope, or kind.
List all connected peer hippo instances.
Mark a memory as retrieved to update its retrieval count and strengthen it.
Record the outcome of a decision or action. Outcomes strengthen or weaken memories based on success/failure, supporting reinforcement learning.
Perform a physics-based search using force-directed graph layout to find memories based on spatial relationships.
Increase the strength of a memory, making it more likely to be retrieved.
Retrieve relevant memories from the project memory store. Returns memories ranked by relevance, strength, and recency within the token budget. Use at session start or when you need context about a topic. Pass include_continuity=true to also surface the active task snapshot, latest matching session handoff, and recent session events as a "## Continuity" appendix.
Store a memory in the project memory store. Memories are automatically strengthened on retrieval and decay over time. Supports tags, scopes, and custom metadata.
Resolve a memory conflict by choosing which version to keep.
Perform a hybrid search combining BM25 keyword matching and semantic similarity.
Share a memory with peer hippo instances.
Run the sleep cycle to trigger consolidation, decay, and other background maintenance tasks.
Mark a memory as superseded by a newer one, reducing its retrieval likelihood.
No permission gates visible on sensitive operations. hippo_forget and hippo_consolidate should verify caller authority before executing. No audit trail or scope declarations.
Pagination not documented for list tools. hippo_list_memories accepts 'limit' but no offset/cursor or total_count return field documented. Baseline: tools returning lists must support pagination.