Can a Fleet hosted agent be configured to hit my Langconnect RAG on GCP?

LangSmith Product Help
1

Can a Fleet hosted agent be configured to hit my Langconnect RAG on GCP ?

no-code abstracted solutions are ‘nice’ but cost of naturally is some inflexibility.. possible to go to a Fleet Agent’s code level and or/ any way to hit my Langconnect RAG on GCP from a Fleet hosted Agent

2

guess only way is to hook to custom MCP server with RAG tool there

3

Hi @johnda98 there isn’t a built-in way to point a Fleet-hosted agent directly at a LangConnect RAG instance on GCP. Fleet’s no-code tools are aimed at common SaaS integrations (Gmail, Slack, BigQuery, etc.), not a self-hosted LangConnect API.

What works today: expose your LangConnect RAG as tools on a custom remote MCP server, then connect that server in Fleet:

  1. Run a small MCP service (on GCP, Cloud Run, etc.) that wraps LangConnect e.g. tools for search, list collections, etc., calling your LangConnect REST endpoints.
  2. Make sure it’s reachable over HTTPS from LangSmith (public URL, or private connectivity if you’re on self-hosted Fleet).
  3. Add auth (API key / bearer token) via MCP headers.
  4. In Fleet: Fleet → Integrations → + Custom MCP, or in the agent editor under Toolbox → MCP. Fleet will discover those tools and call them at runtime.

Docs: Remote MCP servers

You can also use the LangSmith Tool Server to build and host those tools if you want OAuth/auth helpers, but any MCP server that speaks the protocol and is reachable from Fleet should work.

If you need more control: export the agent to Deep Agents code and call LangConnect’s REST API directly from Python/JS, or keep Fleet for orchestration and handle RAG in your own app via the Fleet API.

Networking note: if LangConnect only lives on a private VPC with no ingress, cloud Fleet won’t reach it until you expose HTTPS (e.g. Cloud Run + IAM/API key) or set up connectivity your self-hosted setup supports.

So your instinct is right, custom MCP with a RAG tool is the main no-code path. Happy to help further if you share how LangConnect is exposed (public URL vs internal only).