II.
KnowledgeFabricImpl overview
Reference · liveknowledge-fabric-impl:supermemory.api
Supermemory REST API & MCP Server overview
Cloud REST API (api.supermemory.ai) and MCP Server 4.0 for agent memory operations. REST API provides document ingestion, memory management, profile retrieval, and hybrid search. MCP server (mcp.supermemory.ai/mcp) exposes four tools: addMemory, search, getProjects, whoAmI. Built on Cloudflare Workers with Durable Objects for persistent sessions. SDKs for TypeScript (npm supermemory) and Python (pip supermemory).
Attributes
displayName
Supermemory REST API & MCP Server
agentVersionId
agentVersion:supermemory:current
fabricKind
cloud-api-and-mcp
description
Cloud REST API (api.supermemory.ai) and MCP Server 4.0 for agent
memory operations. REST API provides document ingestion, memory
management, profile retrieval, and hybrid search. MCP server
(mcp.supermemory.ai/mcp) exposes four tools: addMemory, search,
getProjects, whoAmI. Built on Cloudflare Workers with Durable
Objects for persistent sessions. SDKs for TypeScript (npm
supermemory) and Python (pip supermemory).
apiBaseUrl
mcpUrl
authentication
primary
OAuth via /.well-known/oauth-protected-resource
alternative
Bearer token with sm_ prefix
keyEndpoints
- POST /v3/documents -- ingest content (text, URL, PDF, image, video)
- GET /v3/profile -- retrieve user profile (static + dynamic)
- GET /v3/search -- semantic/hybrid search with metadata filtering
- POST /v3/memories -- create/update memories
- DELETE /v3/memories -- soft-delete (forget)
mcpTools
- addMemory -- save information with optional project scoping
- search -- retrieve memories and profiles
- getProjects -- list available projects
- whoAmI -- authenticated user details
sdkUsage
TypeScript: client.add(), client.profile(), client.search.memories()
Python: client.add(), client.profile(), client.search()
Both support containerTag for user/project scoping.
processingPipeline
Queued -> Extracting -> Chunking -> Embedding -> Indexing -> Done
knowledgeGraphRelationships
- Update -- new info contradicts old, tracks isLatest
- Extend -- new info enriches without replacing
- Derive -- system infers connections from patterns
ingestionOptions
taskType
memory (full context layer) or superrag (managed RAG)
dreaming
dynamic (batch related docs) or instant (process immediately)
containerTag
scope to user/project (max 100 chars)
metadata
key-value pairs for filtering
ourEquivalent
Our MCP client can consume Supermemory's MCP server directly by
adding its endpoint to .mcp.json. However, for orchestrated agent
runs, the SMFS approach (supermemory-smfs.yaml) is preferred over
raw MCP tools because it avoids adding cognitive load to the agent
prompt. The REST API is most useful for the orchestrator layer
(genty-platform) to perform structured memory operations like
profile retrieval at run start or batch memory ingestion at run end.
Outgoing edges
integrates_with2
part_of1
- agent:supermemory·AgentProductSupermemory
realizes1
- layer:12-knowledge-fabric·LayerKnowledge Fabric
Incoming edges
None.