Long-term memory shared by Claude Code, Cursor, Codex, Copilot and any MCP client: typed memories, hybrid vector + keyword search, a memory graph, contradiction detection, and task claims so several AI tools don't edit the same files. Hosted or self-hosted.
Scored across 25 tools
Several retrieval tools overlap heavily: recall, recall_multi_session, search_temporal, get_recent_memories, get_timeline, and get_memory_chain all retrieve memories with only subtle differences. The conflict/evolution tools (find_contradictions, mee_evolve, mee_reason, mee_resolve) also have blurred boundaries, though descriptions provide some guidance.
Most tools follow a predictable snake_case verb_noun pattern (claim_task, delete_memory, get_active_work). The mee_* namespace and a few bare or non-verb-led names (recall, remember, pre_action_check) are minor deviations rather than systemic inconsistency.
At 25 tools, the surface is heavy for a memory server, especially with several overlapping retrieval and evolution variants. The added multi-agent coordination tools justify some breadth, but the set feels borderline over-scoped rather than tightly curated.
The memory lifecycle is mostly covered: store, retrieve, search temporally, inspect chains/timelines, export, delete, and evolve/resolve conflicts. The main gap is an explicit update/edit operation for a memory, which agents may have to work around via evolution or replacement.
Claim a file or task with intent to work on it. Claims are shared with agents in the same account/space and expire after the TTL.
Inputs: intent (required), resource (required), ttl.
Delete one memory by ID, e.g. when the user says it is wrong or asks you to forget it.
Inputs: memory_id (required).
Export all memories for the agent.
Inputs: no named parameters.
Summarize contradictions the memory engine has found: how many beliefs were updated by newer memories and how many are still unresolved. Use mee_resolve to settle a specific pair.
Inputs: no named parameters.
Break a longer narrative into atomic facts for better searchability. Automatically extracts entities and relationships.
Inputs: text (required), context, auto_save.
Get coordination context (claims/intents) for a specific resource in the shared account/space.
Inputs: resource (required).
Get ALL active claims and intents for the shared account/space.
Inputs: no named parameters.
Get the knowledge chain for a memory, showing how it evolved. Returns updates (state changes), extends (refinements), and derives (inferences).
Inputs: memory_id (required).
List the most recently saved memories (newest first), e.g. to review what was stored.
Inputs: type, limit.
Check the health and status of the Memro MCP server and its connections.
Inputs: no named parameters.
Get statistics about your memories: totals for today, this week and this month, average days between memories, and the count per memory type.
Inputs: no named parameters.
Get the full historical timeline of a memory, including all its versions and updates.
Inputs: memory_id (required).
Atomically summarizes work and releases a resource claim for the next agent.
Inputs: summary (required), resource (required), next_agent_id.
Extend an active lease on a resource so it does not expire.
Inputs: resource (required), ttl.
Declare your non-blocking intention for a resource (shared with agents in the same account/space).
Inputs: intent (required), resource (required), ttl.
Trigger a full evolution cycle (decay, consolidation, conflict detection) for the current agent.
Inputs: no named parameters.
Trigger the Memory Evolution Engine to reason about a specific query using its internal knowledge graph.
Inputs: query (required).
Resolve a detected conflict between two memories using evidence-based scoring.
Inputs: memory_a_id (required), memory_b_id (required).
Check for conflicts or overlaps before performing an action against shared leases and intents.
Inputs: intent (required), resource (required).
Search the user's long-term memory. Call it at the start of a task and before answering questions about earlier work, preferences or decisions.
Inputs: query (required), type, limit.
Recall memories across multiple sessions with synthesis. Connects related facts from different conversations and resolves contradictions.
Inputs: query (required), limit, include_all_sessions.
Release a claimed resource and log outcome. Releases the lease in the shared account/space.
Inputs: resource (required), status.
Save something worth knowing in future sessions: a preference, decision, correction, or lasting fact about the user or their project. One short, self-contained sentence per memory. Never store secrets.
Inputs: content (required), type, is_atomic, event_date, +2 more.
Store a new memory with rich metadata context (file, git, function).
Inputs: content (required), type, event_date, session_id, +1 more.
Search memories with temporal reasoning. Finds relevant memories considering both when the conversation happened and when events actually occurred. Perfect for multi-session synthesis and temporal queries.
Inputs: query (required), limit, after_date, atomic_only, +2 more.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Add one secure layer between your agents and this server.