MCP Toolbox for Databases in n8n no bridge needed — it speaks HTTP natively.
Most stdio MCP servers need a gateway before n8n can use them. Google's MCP Toolbox for Databases doesn't: it serves MCP over HTTP out of the box, so the MCP Client Tool node connects straight to it.
MCP Toolbox for Databases (github.com/googleapis/genai-toolbox, 15.5k stars) is Google's open-source MCP server for databases: you declare sources and named, parameterized query tools in a tools.yaml, and agents get that tool set — nothing more. Supported engines include PostgreSQL, MySQL, SQL Server, BigQuery, and Spanner.
n8n has no MCP config file; you attach servers per-workflow through the MCP Client Tool node, which expects an SSE or streamable HTTP endpoint. That requirement usually forces a stdio-to-HTTP bridge — but the Toolbox is one of the rare servers that listens on HTTP natively, making it an unusually clean n8n citizen.
Start the Toolbox as an HTTP service
Run the container with your tools.yaml mounted, --address 0.0.0.0 so the listener is reachable through Docker's port mapping, and port 5000 published. The MCP endpoint is /mcp on that port.
docker run -d --name toolbox -p 5000:5000 \
-v /abs/path/tools.yaml:/config/tools.yaml \
us-central1-docker.pkg.dev/database-toolbox/toolbox/toolbox:latest \
--tools-file /config/tools.yaml --address 0.0.0.0
# MCP Client Tool node settings:
# Endpoint: http://localhost:5000/mcp
# (http://host.docker.internal:5000/mcp if n8n itself runs in Docker)Configure the MCP Client Tool node
In your workflow, add an MCP Client Tool node and connect it to an AI Agent node's tool input. Set the endpoint to the Toolbox URL and pick the streamable HTTP option. Once the node connects, the tools you declared in tools.yaml become callable by the agent, and you can restrict which ones are exposed in the node's settings — sensible if a workflow should only ever read.
From there it's ordinary n8n: the agent decides mid-run when a step needs the database, calls the named tool with parameters, and the result flows back into the conversation.
A good fit for unattended workflows
n8n workflows often run on schedules or webhooks with nobody watching, which is exactly when you don't want an agent improvising SQL. The Toolbox's contract — only the queries in tools.yaml exist — bounds what an unattended run can do to your data, and every callable operation is reviewable in one file.
Troubleshooting in n8n
Connection failures are nearly always Docker networking. If n8n runs in its own container, localhost points at that container, not your host — use host.docker.internal (Mac/Windows) or put both containers on a shared Docker network and use the service name. If the endpoint times out from everywhere, you likely launched the Toolbox without --address 0.0.0.0, so it's listening on loopback inside its container where the port mapping can't reach it.
When the node connects but lists no tools, check docker logs toolbox: the Toolbox reports at startup which tool definitions validated and whether the database source connected. n8n itself rarely needs restarting — fix the container, then re-test the node connection. For n8n Cloud, remember its servers can't reach your private network at all; the Toolbox endpoint must be reachable from the internet, with auth in front, before a cloud workflow can use it.
The live listing
Check the Toolbox's live-probed tool list on its Loomal marketplace listing at https://loomal.ai/marketplace/mcp-toolbox-for-databases before you wire up a workflow. Maintainers of MCP servers can claim listings like this one to verify ownership and set per-call pricing for paying agents.
FAQ
How do I install the Toolbox for Databases in n8n?
Run the Toolbox container with your tools.yaml mounted, --address 0.0.0.0, and port 5000 published, then point an MCP Client Tool node at http://localhost:5000/mcp and attach it to an AI Agent node. No stdio bridge is needed because the Toolbox serves MCP over HTTP natively.
Where is n8n's MCP config file?
There isn't one — MCP servers are configured inside workflows via the MCP Client Tool node's settings (endpoint URL plus optional credentials), not in a file on disk. Each workflow declares the servers it uses.
Why does the node fail to connect to the Toolbox?
Docker networking, almost every time. n8n-in-Docker can't see your host's localhost (use host.docker.internal or a shared network), and a Toolbox started without --address 0.0.0.0 listens where the port mapping can't reach. docker logs toolbox confirms whether the server even started cleanly.
Can I use this from n8n Cloud?
Only if the Toolbox endpoint is reachable from the public internet, since n8n Cloud can't dial into your private network. If you expose it, put authentication in front — this endpoint executes queries against your database.
Browse more MCP servers for n8n.
Endpoint-ready servers with live-probed tool lists.