memanto-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| MOORCHEH_API_KEY | Yes | Your Moorcheh API key. Sign up at https://console.moorcheh.ai/api-keys. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| memanto_rememberA | Store a typed memory for an agent. Args: agent_id: Which agent's memory to write to (must already exist; create with memanto_create_agent). content: The memory content (1–10000 chars). memory_type: One of fact, preference, goal, decision, artifact, learning, event, instruction, relationship, context, observation, commitment, error. title: Optional short title; auto-derived from content if omitted. tags: Optional list of free-form tags for filtering on recall. confidence: Confidence score in [0.0, 1.0]; defaults to 0.8. source: Free-form source label (e.g. "claude", "user", a tool name). provenance: How the memory was obtained — explicit_statement, inferred, observed, validated, corrected, imported. Returns: dict with the stored memory's id and namespace. |
| memanto_recallB | Semantic search over an agent's memory. Args: query: Natural-language query. agent_id: Whose memory to search. memory_types: Optional filter by type(s). tags: Optional filter by tag(s). limit: Max results (default 10). min_confidence: Optional minimum confidence threshold. Returns: dict with search hits (content, score, metadata). |
| memanto_recall_as_ofC | Recall memories as they existed at a specific date. Args: as_of_date: ISO 8601 date or datetime (e.g. '2026-01-15' or '2026-01-15T12:00:00Z'). |
| memanto_recall_currentC | Recall only currently-active memories (excludes superseded, expired, deleted). |
| memanto_answerB | RAG-grounded answer using the agent's memory. Memanto retrieves relevant memories and synthesizes an answer with citations. Use this when the user wants a synthesized response, not raw search hits. |
| memanto_create_agentA | Create a new agent (provisions a dedicated Moorcheh namespace). Args: agent_id: Unique kebab-case identifier (e.g. 'customer-support', 'geoff-personal'). description: Optional human description. pattern: One of support, project, tool. Default 'support'. |
| memanto_list_agentsA | List all agents this MCP server knows about. |
| memanto_get_agentB | Get metadata for a single agent (or None if it doesn't exist). |
| memanto_delete_agentA | Delete an agent and its memory namespace. DESTRUCTIVE: this removes the agent's stored memories. Confirm with the user before calling. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 9 tools
The forget/remember/recall core is clear, but three retrieval tools (recall, recall_as_of, recall_current) overlap: the docs never state whether plain `recall` already excludes superseded memories, and recall_as_of/recall_current document no agent_id or filter args, so an agent cannot tell whether they are scoped per-agent or global. The remember-vs-answer split (raw storage vs synthesized RAG) is well delineated.
Every tool uses the same `memanto_` prefix with a snake_case verb or verb_object form (remember, recall, answer, create_agent, list_agents, get_agent, delete_agent). The recall_* variants form a predictable modifier pattern rather than a convention break.
Nine tools is well-scoped: five memory operations plus four agent lifecycle operations, each earning its place. No obvious filler tools or redundancies that would bloat the surface.
Agent management is full CRUD except update_agent, but the memory lifecycle is write-and-read only — there is no way to update, forget, delete, or supersede an individual memory, despite memory types like 'error' and 'commitment' implying correction workflows. recall_as_of and recall_current also lack documented parameters, leaving dead ends for filtered use.