[Feature] Allow configuring Neo4j database name (multi-DB support)

[MumuTW]

🗂️ Background

I’m trying to run graphiti_mcp_server.py with a custom database hosted in a Neo4j instance.
Currently, Neo4jConfig only reads NEO4J_URI, NEO4J_USER, and NEO4J_PASSWORD from environment variables.
There is no option to specify a target database, so the server always connects to the default neo4j database.

✅ Expectation

Users should be able to set NEO4J_DATABASE=my_custom_db in .env to configure which target database to use. All Neo4jDriver sessions should run Cypher queries against that specific database instead of silently defaulting to the built-in neo4j database.

⚡Actual Behavior

• Neo4jConfig correctly reads NEO4J_URI, NEO4J_USER, and NEO4J_PASSWORD — but ignores NEO4J_DATABASE.
• Neo4jDriver sets self._database to 'neo4j' by default at init, so all sessions fallback to the default DB.
• No place in the current Graphiti or Neo4jDriver logic explicitly overrides the target DB for each session() or execute_query().
• Even if you provide NEO4J_DATABASE in .env, it does nothing — causing unexpected cross-environment data mixing or permission errors if the user doesn’t have access to the default neo4j DB.

🔑 Why This Matters

• Neo4j supports multiple isolated databases since 4.x.
• Real-world teams rely on multi-tenancy: staging, production, test DBs, or even tenant-per-graph setups.
• People expect full DB routing to be configured declaratively via .env, without patching Python code.
• Fallback to 'neo4j' is fine if not set — but users must be able to explicitly override the DB safely.

🧩 Minimal repro

# .env
NEO4J_URI=bolt://localhost:7687
NEO4J_USER=my-user
NEO4J_PASSWORD=my-pass
NEO4J_DATABASE=my-database

# run
uv run graphiti_mcp_server.py --model gpt-4.1-mini

✅ Connection to bolt://localhost:7687 works, ❌ But queries fail if the user doesn’t have access to the default neo4j DB.

🕵️‍♂️ Root Cause

• Neo4jDriver currently pins self._database to 'neo4j' by default.
• session() and execute_query() use this pinned value if no database is passed.
• Graphiti does not propagate database=... per request, so it always falls back to the pinned value.
• There is no config parsing for NEO4J_DATABASE.

📝Proposed solution

• [ ] Add database: str to Neo4jConfig and parse NEO4J_DATABASE from the environment, with fallback to 'neo4j'.

• [ ] Update Neo4jDriver:

def __init__(..., database: str = 'neo4j'):
    self._database = database

• [ ] Pass database to all session() and execute_query() calls:

def session(self, database: str | None = None):
    return self.client.session(database=database or self._database)

async def execute_query(self, cypher_query_, database: str | None = None, ...):
    return await self.client.execute_query(..., database_=database or self._database)

• [ ] Update Graphiti to store self.database and always pass it when opening sessions.
• [ ] Document NEO4J_DATABASE usage in .env.example and README.
• [ ] Consider adding an integration test for multi-DB routing.

🙏 Extra context

Happy to open a PR for this if you’d like — please confirm if there’s any design constraint I should know about! Reference: Neo4j Python Driver supports multi-database connections since 4.x: https://neo4j.com/docs/api/python-driver/current/api.html#session

Thanks for making Graphiti awesome! 🚀✨

[jeanlucthumm]

graphiti seems to have a bunch of possible drivers outside of Neo4j that would have to take the top level db param. Not sure what to do about KuzuDriver and NeptuneDriver.

[jeanlucthumm]

FYI: Neo4j community edition does not support multiple databases, it's always neo4j. This is an Enterprise feature.

https://neo4j.com/docs/upgrade-migration-guide/current/version-4/migration/drivers/multiple-db/

[claude[bot]]

@MumuTW Is this still an issue? Please confirm within 14 days or this issue will be closed.

[claude[bot]]

@MumuTW Is this still relevant? Please confirm within 14 days or this issue will be closed.