Synapse Layer — Trust Infrastructure for AI Agents
MCP-native Trust Infrastructure for AI Agents. Persistent encrypted memory with Trust Quotient.
我該用這個嗎
品質與安全性
A
發現項目（4）
- LOWTool 'save_to_synapse' description lacks action verb在 save_to_synapse 中
- LOWTool 'store_memory' description lacks action verb在 store_memory 中
- LOWTool 'neural_handover' description lacks action verb在 neural_handover 中
- LOWTool 'slo_report' description lacks action verb在 slo_report 中
根據工具定義與協定合規性的自動化分析。
上下文成本
~3,950Token（工具定義）
~3.2 KB典型回應大小
顯著的注意力影響（128k 上下文的 3.09%）
這是每次將伺服器的工具載入模型上下文時所消耗的約略 token 數量。數量越高，可用於其他工作的注意力就越少。
安裝
一鍵安裝
將以下內容加入你的 `claude_desktop_config.json` 檔案：
{
"mcpServers": {
"synapse-layer": {
"url": "https://forge.synapselayer.org/api/mcp"
}
}
}遠端端點
https://forge.synapselayer.org/api/mcpstreamable-http它能做什麼
工具清單
工具（13）
🟢 唯讀🟡 寫入🔴 刪除⚪ 未知
🟢recall(query, reason, scope, agent_id, limit, ...)
Retrieve relevant persisted memory using semantic, temporal, priority, or hybrid routing. Governance: requires reason (10–200 chars). Rate limit: 20/min.
輸入結構描述
{
"type": "object",
"properties": {
"query": {
"type": "string",
"description": "What to recall — natural language query for memory retrieval."
},
"reason": {
"type": "string",
"description": "REQUIRED (10–200 chars). Audit policy: MEMORY_EXPORT_GOVERNANCE_V1. Human-readable justification for retrieving PLAINTEXT memory. Without this parameter the call WILL fail with code -32602. Example: \"User asked me to summarize last week decisions.\""
},
"scope": {
"type": "string",
"enum": [
"agent",
"tenant"
],
"description": "Explicit memory scope. \"agent\" (default, fail-closed) = limits recall to the given agent_id (or the token agent). \"tenant\" = cross-agent recall within the authenticated tenant. If omitted with agent_id present, \"agent\" is inferred; if both omitted, the token agent is used (verified-token scope)."
},
"agent_id": {
"type": "string",
"description": "Agent identifier (canonical or alias). Used only when scope=\"agent\" or scope is inferred."
},
"limit": {
"type": "number",
"description": "Maximum memories to return (1–50, default: 10)."
},
"mode": {
"type": "string",
"enum": [
"auto",
"temporal",
"semantic",
"priority",
"hybrid"
],
"description": "Recall routing mode. \"temporal\" = pure chronological, \"semantic\" = FTS + TQ ranking, \"priority\" = critical/high-TQ first, \"hybrid\" = weighted blend, \"auto\" = detect from query (default)."
},
"subject": {
"type": "string",
"description": "ESCOPO1 (optional): opaque 8-128 char token ([A-Za-z0-9._:-]) that isolates memory to a single person within a shared connect-token. Omit for token-scoped behavior. Never include PII/email."
}
},
"required": [
"query",
"reason"
]
}輸出結構描述
{
"type": "object",
"properties": {
"memories": {
"type": "array",
"items": {
"type": "object",
"properties": {
"id": {
"type": "string"
},
"content": {
"type": "string"
},
"similarity": {
"type": "number"
},
"createdAt": {
"type": "string",
"format": "date-time"
}
}
}
},
"count": {
"type": "number"
},
"scope": {
"type": "string",
"enum": [
"agent",
"tenant"
]
}
},
"required": [
"memories",
"count",
"scope"
]
}🟡save_to_synapse(content, agent_id, project, type, importance, ...)
Persist memory with encryption at rest, sanitization, and deduplication controls.
輸入結構描述
{
"type": "object",
"properties": {
"content": {
"type": "string",
"description": "The memory content to store securely."
},
"agent_id": {
"type": "string",
"description": "Agent identifier for memory isolation. Defaults to \"default\"."
},
"project": {
"type": "string",
"description": "Project identifier (e.g., SYNAPSE_LAYER)."
},
"type": {
"type": "string",
"description": "Event type: [MILESTONE], [DECISION], [ALERT], [AUTO-STRAT], [AUTO-OP], [AUTO-INSIGHT], [AUTO-DECISION], [AUTO-CONTEXT], [MANUAL]."
},
"importance": {
"type": "number",
"description": "Accepted for compatibility; not enforced by the API."
},
"tags": {
"type": "array",
"items": {
"type": "string"
},
"description": "Tags for categorization."
},
"force_intent": {
"type": "string",
"description": "MANUAL OVERRIDE ONLY — use apenas quando você (humano) quer forçar explicitamente a classificação de intenção. Valores reconhecidos: preference | fact | procedural | bio | critical | operational | strategic | compliance | security | system_directive | clinical | general | ephemeral. Se não tiver certeza, NÃO preencha este campo — o sistema classifica automaticamente. Valores não reconhecidos são ignorados silenciosamente e retornados em warnings."
},
"intent_reason": {
"type": "string",
"description": "Auditable reason for forcing intent classification. Max 500 characters. Ignored if force_intent is absent.",
"maxLength": 500
},
"subject": {
"type": "string",
"description": "ESCOPO1 (optional): opaque 8-128 char token ([A-Za-z0-9._:-]) that isolates memory to a single person within a shared connect-token. Omit for token-scoped behavior. Never include PII/email."
}
},
"required": [
"content"
]
}輸出結構描述
{
"type": "object",
"properties": {
"memory_id": {
"type": "string"
},
"status": {
"type": "string"
},
"trust_quotient": {
"type": "number"
}
},
"required": [
"memory_id",
"status"
]
}🟢process_text(text, agent_id, project, source)
Extract candidate memories from free-form text with governance filters and sanitization.
輸入結構描述
{
"type": "object",
"properties": {
"text": {
"type": "string",
"description": "Free-form text to scan for auto-save triggers."
},
"agent_id": {
"type": "string",
"description": "Agent identifier. Defaults to \"default\"."
},
"project": {
"type": "string",
"description": "Force a specific project (e.g., SYNAPSE_LAYER, OFFLY). Auto-detected if omitted."
},
"source": {
"type": "string",
"description": "Source identifier (default: mcp)."
}
},
"required": [
"text"
]
}輸出結構描述
{
"type": "object",
"properties": {
"extracted": {
"type": "array",
"items": {
"type": "object"
}
},
"storedCount": {
"type": "number"
}
},
"required": [
"extracted",
"storedCount"
]
}🟢search(query, reason, scope, agent_id, limit, ...)
Search persisted memory across agent scopes using full-text matching. Governance: requires reason (10–200 chars) because results contain plaintext content.
輸入結構描述
{
"type": "object",
"properties": {
"query": {
"type": "string",
"description": "Search query — natural language or keywords."
},
"reason": {
"type": "string",
"description": "REQUIRED (10–200 chars). Audit policy: MEMORY_EXPORT_GOVERNANCE_V1. Human-readable justification for retrieving PLAINTEXT memory. Without this parameter the call WILL fail with code -32602. Example: \"Checking prior migration decisions for the user.\""
},
"scope": {
"type": "string",
"enum": [
"agent",
"tenant"
],
"description": "Explicit search scope. \"agent\" (default, fail-closed) = limits search to the given agent_id (or the token agent). \"tenant\" = cross-agent search within the authenticated tenant. If omitted with agent_id present, \"agent\" is inferred; if both omitted, the token agent is used (verified-token scope)."
},
"agent_id": {
"type": "string",
"description": "Agent identifier (canonical or alias). Used only when scope=\"agent\" or scope is inferred."
},
"limit": {
"type": "number",
"description": "Maximum results to return (1–50, default: 20)."
},
"subject": {
"type": "string",
"description": "ESCOPO1 (optional): opaque 8-128 char token ([A-Za-z0-9._:-]) that isolates memory to a single person within a shared connect-token. Omit for token-scoped behavior. Never include PII/email."
}
},
"required": [
"query",
"reason"
]
}輸出結構描述
{
"type": "object",
"properties": {
"results": {
"type": "array",
"items": {
"type": "object"
}
},
"count": {
"type": "number"
},
"scope": {
"type": "string",
"enum": [
"agent",
"tenant"
]
}
},
"required": [
"results",
"count",
"scope"
]
}🟢health_check
Check service availability, engine version, and storage health. Public method.
輸入結構描述
{
"type": "object",
"properties": {}
}輸出結構描述
{
"type": "object",
"properties": {
"status": {
"type": "string"
},
"version": {
"type": "string"
},
"dbLatencyMs": {
"type": "number"
}
},
"required": [
"status",
"version"
]
}🟢initialize_context(detail, reason)
Initialize a persistent memory context for a conversation or agent session. Returns REDACTED metadata by default (detail="summary"). Use detail="full" with a reason (10–200 chars) to retrieve plaintext content under MEMORY_EXPORT_GOVERNANCE_V1.
輸入結構描述
{
"type": "object",
"properties": {
"detail": {
"type": "string",
"enum": [
"summary",
"full"
],
"description": "Detail level. \"summary\" (default) = redacted metadata only, no plaintext. \"full\" = plaintext content; REQUIRES the reason parameter and is audit-logged."
},
"reason": {
"type": "string",
"description": "REQUIRED only when detail=\"full\" (10–200 chars). Audit policy: MEMORY_EXPORT_GOVERNANCE_V1. Human-readable justification for retrieving PLAINTEXT context. Without it, detail=\"full\" WILL fail with code -32602."
}
},
"required": []
}輸出結構描述
{
"type": "object",
"properties": {
"memories": {
"type": "array",
"items": {
"type": "object"
}
},
"scope": {
"type": "string",
"enum": [
"agent",
"tenant"
]
},
"agent_id": {
"type": "string"
}
},
"required": [
"scope"
]
}🟡save_memory(content, agent_id, tags, force_intent, intent_reason, ...)
Save a memory entry to the persistent store. Alias of save_to_synapse.
輸入結構描述
{
"type": "object",
"properties": {
"content": {
"type": "string",
"description": "The memory content to store securely."
},
"agent_id": {
"type": "string",
"description": "Agent identifier for memory isolation."
},
"tags": {
"type": "array",
"items": {
"type": "string"
},
"description": "Tags for categorization."
},
"force_intent": {
"type": "string",
"description": "MANUAL OVERRIDE ONLY — use apenas quando você (humano) quer forçar explicitamente a classificação de intenção. Valores reconhecidos: preference | fact | procedural | bio | critical | operational | strategic | compliance | security | system_directive | clinical | general | ephemeral. Se não tiver certeza, NÃO preencha este campo — o sistema classifica automaticamente. Valores não reconhecidos são ignorados silenciosamente e retornados em warnings."
},
"intent_reason": {
"type": "string",
"description": "Auditable reason for forcing intent classification. Max 500 characters. Ignored if force_intent is absent.",
"maxLength": 500
},
"subject": {
"type": "string",
"description": "ESCOPO1 (optional): opaque 8-128 char token ([A-Za-z0-9._:-]) that isolates memory to a single person within a shared connect-token. Omit for token-scoped behavior. Never include PII/email."
}
},
"required": [
"content"
]
}輸出結構描述
{
"type": "object",
"properties": {
"id": {
"type": "string"
},
"stored": {
"type": "boolean"
},
"scope": {
"type": "string",
"enum": [
"agent",
"tenant"
]
}
},
"required": [
"id",
"stored",
"scope"
]
}⚪store_memory(content, agent_id, tags, force_intent, intent_reason, ...)
Store structured memory with metadata and trust scoring.
輸入結構描述
{
"type": "object",
"properties": {
"content": {
"type": "string",
"description": "The memory content to store securely."
},
"agent_id": {
"type": "string",
"description": "Agent identifier for memory isolation. Defaults to \"default\"."
},
"tags": {
"type": "array",
"items": {
"type": "string"
},
"description": "Tags for categorization."
},
"force_intent": {
"type": "string",
"description": "MANUAL OVERRIDE ONLY — use apenas quando você (humano) quer forçar explicitamente a classificação de intenção. Valores reconhecidos: preference | fact | procedural | bio | critical | operational | strategic | compliance | security | system_directive | clinical | general | ephemeral. Se não tiver certeza, NÃO preencha este campo — o sistema classifica automaticamente. Valores não reconhecidos são ignorados silenciosamente e retornados em warnings."
},
"intent_reason": {
"type": "string",
"description": "Auditable reason for forcing intent classification. Max 500 characters. Ignored if force_intent is absent.",
"maxLength": 500
},
"subject": {
"type": "string",
"description": "ESCOPO1 (optional): opaque 8-128 char token ([A-Za-z0-9._:-]) that isolates memory to a single person within a shared connect-token. Omit for token-scoped behavior. Never include PII/email."
}
},
"required": [
"content"
]
}輸出結構描述
{
"type": "object",
"properties": {
"id": {
"type": "string"
},
"stored": {
"type": "boolean"
},
"scope": {
"type": "string",
"enum": [
"agent",
"tenant"
]
}
},
"required": [
"id",
"stored",
"scope"
]
}🟢recall_memory(query, reason, scope, agent_id, limit, ...)
Recall persisted memory by query. Alias of recall.
輸入結構描述
{
"type": "object",
"properties": {
"query": {
"type": "string",
"description": "What to recall — natural language query for memory retrieval."
},
"reason": {
"type": "string",
"description": "REQUIRED (10–200 chars). Audit policy: MEMORY_EXPORT_GOVERNANCE_V1. Without this the call WILL fail with code -32602. Example: \"Checking user project status.\""
},
"scope": {
"type": "string",
"enum": [
"agent",
"tenant"
],
"description": "Explicit memory scope. \"agent\" (default, fail-closed) = limits recall to the given agent_id (or the token agent). \"tenant\" = cross-agent recall within the authenticated tenant. If omitted with agent_id present, \"agent\" is inferred; if both omitted, the token agent is used (verified-token scope)."
},
"agent_id": {
"type": "string",
"description": "Agent identifier (canonical or alias). Used only when scope=\"agent\" or scope is inferred."
},
"limit": {
"type": "number",
"description": "Maximum memories to return (1–50, default: 10)."
},
"mode": {
"type": "string",
"enum": [
"auto",
"temporal",
"semantic",
"priority",
"hybrid"
],
"description": "Recall routing mode."
},
"subject": {
"type": "string",
"description": "ESCOPO1 (optional): opaque 8-128 char token ([A-Za-z0-9._:-]) that isolates memory to a single person within a shared connect-token. Omit for token-scoped behavior. Never include PII/email."
}
},
"required": [
"query",
"reason"
]
}輸出結構描述
{
"type": "object",
"properties": {
"memories": {
"type": "array",
"items": {
"type": "object",
"properties": {
"id": {
"type": "string"
},
"content": {
"type": "string"
},
"similarity": {
"type": "number"
},
"createdAt": {
"type": "string",
"format": "date-time"
}
}
}
},
"count": {
"type": "number"
},
"scope": {
"type": "string",
"enum": [
"agent",
"tenant"
]
}
},
"required": [
"memories",
"count",
"scope"
]
}🟢list_memories(scope, agent_id, limit, subject)
List memory metadata with pagination and governance limits.
輸入結構描述
{
"type": "object",
"properties": {
"scope": {
"type": "string",
"enum": [
"agent",
"tenant"
],
"description": "Explicit memory scope. \"agent\" (default, fail-closed) = lists memories for the given agent_id (or the token agent). \"tenant\" = lists memories across all agents of the authenticated tenant."
},
"agent_id": {
"type": "string",
"description": "Agent identifier (canonical or alias). Used only when scope=\"agent\" or scope is inferred."
},
"limit": {
"type": "number",
"description": "Maximum memory metadata rows to return. Hard-capped: default 5, max 10. Values above 10 are silently reduced to 10."
},
"subject": {
"type": "string",
"description": "ESCOPO1 (optional): opaque 8-128 char token ([A-Za-z0-9._:-]) that isolates memory to a single person within a shared connect-token. Omit for token-scoped behavior. Never include PII/email."
}
}
}輸出結構描述
{
"type": "object",
"properties": {
"memories": {
"type": "array",
"items": {
"type": "object"
}
},
"count": {
"type": "number"
},
"total": {
"type": "number"
}
},
"required": [
"memories",
"count"
]
}🟡memory_feedback(memory_id, memoryId, signal, session_id, sessionId)
Submit feedback to adjust trust scoring for stored memories.
輸入結構描述
{
"type": "object",
"properties": {
"memory_id": {
"type": "string",
"description": "ID of the memory to provide feedback on (canonical snake_case name)."
},
"memoryId": {
"type": "string",
"description": "Deprecated camelCase alias of memory_id — still accepted for backward compatibility."
},
"signal": {
"type": "string",
"enum": [
"used",
"ignored",
"helpful",
"irrelevant"
],
"description": "Feedback signal: used (auto), helpful (positive), irrelevant (negative), ignored."
},
"session_id": {
"type": "string",
"description": "Optional session identifier for tracking (canonical snake_case name)."
},
"sessionId": {
"type": "string",
"description": "Deprecated camelCase alias of session_id — still accepted for backward compatibility."
}
},
"required": [
"signal"
]
}輸出結構描述
{
"type": "object",
"properties": {
"memoryId": {
"type": "string"
},
"success": {
"type": "boolean"
},
"signal": {
"type": "string"
}
},
"required": [
"memoryId",
"success"
]
}⚪neural_handover(token, reason, consuming_agent)
Transfer contextual state between agents with continuity controls.
輸入結構描述
{
"type": "object",
"properties": {
"token": {
"type": "string",
"description": "The handover token (64 hex chars). Generated by the source agent/user in Forge UI."
},
"reason": {
"type": "string",
"description": "REQUIRED (10–200 chars). Human-readable justification for consuming this handover. Example: \"Receiving project context from Claude session.\""
},
"consuming_agent": {
"type": "string",
"description": "Agent identifier consuming the handover (defaults to token agent)."
}
},
"required": [
"token",
"reason"
]
}輸出結構描述
{
"type": "object",
"properties": {
"handoverId": {
"type": "string"
},
"status": {
"type": "string"
}
},
"required": [
"status"
]
}🟢slo_report(admin_token, window_hours)
Return uptime and SLO metrics for the MCP service.
輸入結構描述
{
"type": "object",
"properties": {
"admin_token": {
"type": "string",
"description": "Admin authentication token. Must match ADMIN_TOKEN environment variable."
},
"window_hours": {
"type": "array",
"items": {
"type": "number"
},
"description": "Time windows in hours for the report. Default: [24, 168] (24h + 7d)."
}
},
"required": [
"admin_token"
]
}輸出結構描述
{
"type": "object",
"properties": {
"window": {
"type": "string"
},
"uptime": {
"type": "number"
},
"p50": {
"type": "number"
},
"p95": {
"type": "number"
},
"errorRate": {
"type": "number"
}
},
"required": [
"uptime"
]
}社群
證據
近期觀測
已驗證未記錄版本13 個工具
已驗證未記錄版本13 個工具