MemoryDB Python client

superb-memory gives you the Python client, the mdb command-line tool and the memorydb-mcp stdio server for MemoryDB — a single-binary Go document database that also stores AI-agent memory (facts with evidence, summaries, tool traces, hybrid BM25 + vector search).

One line to a running database

pip install superb-memory
mdb serve                    # fetches the server binary, starts it on 127.0.0.1:8080

mdb serve prints the connection string; every client reads it from the environment:

export MEMORYDB_URL=memorydb://TOKEN@127.0.0.1:8080

Documents

from memorydb import MemoryDB

db = MemoryDB()                       # reads MEMORYDB_URL
db.collections.create("customers",
                      fields={"email": {"type": "string", "required": True},
                              "age": {"type": "integer"}},
                      unique=["email"], indexes=["age"])

customers = db.collection("customers")
row = customers.objects.create(email="demo@example.com", age=24, idempotency_key="customer-demo-1")
customers.objects.filter(age__gte=18).order_by("-age").only("email").all()
customers.objects.increment(pk=row["id"], views=1)     # atomic counter, no version needed

Django-style lookups: eq ne gt gte lt lte in contains icontains prefix iequals exists, plus or_q, Sum/Avg/Min/Max with group_by, cursor pagination and optimistic version checks.

Agent memory

db.ingest([{"role": "user", "content": "I prefer terse answers in Hindi"}],
          facts=[{"subject": "user", "predicate": "answer_style", "value": "terse Hindi",
                  "quote": "terse answers in Hindi", "message_index": 0, "confidence": 0.95}])
db.objects.search("terse")

Facts need an exact quote from a user message, so a stored memory always carries its evidence.

CLI and MCP

mdb health; mdb search "fridays"; mdb memory add --content "user ships on fridays"
memorydb-mcp          # stdio MCP server for Claude Desktop, Qoder, IDE agents

Companion packages: memorydb-django (cache + session backends) and memorydb-langchain (chat message history). Docs: docs/QUICKSTART.md and docs/CONNECTING.md in the repository.