MCP Tools Reference

Complete reference for all 13 tools exposed by the Synapse Layer MCP server. Each tool includes input schema, output schema, and behavioral annotations.

On this page

Store Tools

Three aliases for storing memories. All pass through the security pipeline and encrypt content server-side with AES-256-GCM.

store_memory

Store structured memory with metadata and trust scoring.

Required Parameters

  • content (string) — Memory content to store

Optional Parameters

  • agent_id (string) — Agent identifier. Default: default
  • tags (string[]) — Categorization tags

save_to_synapse

Persist memory with encryption, sanitization, and deduplication controls.

Required Parameters

  • content (string) — Memory content to store securely

Optional Parameters

  • agent_id (string) — Agent identifier
  • project (string) — Project identifier
  • type (string) — Event type: MILESTONE, DECISION, ALERT, AUTO-STRAT, AUTO-OP, AUTO-INSIGHT
  • importance (number) — Importance level 1–5 (default: 3)
  • tags (string[]) — Categorization tags

save_memory

Alias of save_to_synapse. Same parameters.

Recall Tools

recall / recall_memory

Retrieve relevant memories using semantic, temporal, priority, or hybrid routing.

Required Parameters

  • query (string) — Natural language query for memory retrieval
  • reason (string) — Justification for the recall request (10–200 chars)

Optional Parameters

  • scope (string) — agent (default) or tenant for cross-agent recall
  • agent_id (string) — Agent identifier for scoped recall
  • limit (number) — Max memories to return (1–50, default: 10)
  • mode (string) — auto | temporal | semantic | priority | hybrid

Search & List

search

Full-text search across persisted memory with scope control.

  • query (string, required) — Search query
  • scope (string) — agent or tenant
  • limit (number) — Max results (1–50, default: 20)

list_memories

List memory metadata with pagination and governance limits.

  • scope (string) — agent or tenant
  • limit (number) — Max rows (default: 5, max: 10)

Process Text

process_text

Extract candidate memories from free-form text with governance filters and sanitization.

  • text (string, required) — Free-form text to scan
  • agent_id (string) — Agent identifier
  • project (string) — Project identifier
  • source (string) — Source identifier (default: mcp)

Feedback & Handover

memory_feedback

Submit feedback to adjust trust scoring for stored memories.

  • memoryId (string, required) — ID of the memory
  • signal (string, required) — used | ignored | helpful | irrelevant
  • sessionId (string) — Optional session identifier

neural_handover

Warning

Neural Handover is in beta. Interface may change.

Transfer contextual state between agents with continuity controls.

  • token (string, required) — Handover token (64 hex chars). Generated in Forge UI or API
  • reason (string, required) — Justification for consuming the handover (10–200 chars)
  • consuming_agent (string) — Agent identifier consuming the handover

System Tools

health_check

Check service availability, engine version, and storage health. No parameters required.

initialize_context

Initialize a persistent memory context for a conversation or agent session. No parameters required.

slo_report

Return uptime and SLO metrics. Admin-only.

  • admin_token (string, required) — Admin authentication token
  • window_hours (number[]) — Time windows (default: [24, 168])

Tool Annotations

Each tool carries MCP annotations that inform clients about behavior:

AnnotationMeaning
readOnlyHint: trueTool does not modify state (recall, search, health_check)
destructiveHint: trueTool may cause irreversible changes (none currently)
idempotentHint: trueCalling twice produces same result (recall, health_check)
openWorldHint: trueTool interacts with external state (memory_feedback, neural_handover)

Next Steps

Server Setup

Connect via Streamable HTTP or stdio.

Claude Desktop

Step-by-step integration guide.

REST API

Same capabilities via HTTP endpoints.