Deploying langflow at scale

[rijojoseph07]

Hi all,

I’m trying to deploy Langflow in production on Kubernetes using the official Helm chart. I’m running into several issues, especially when deploying the backend and frontend separately:

• Job queue not found: The backend frequently fails with messages about the job queue being missing.
• Ephemeral MCP server configs: The MCP configurations are currently stored in local files. This causes problems because when a request lands on a new pod (due to scaling or restarts), newly added MCP servers vanish or are unavailable; it looks like there’s no persistence or centralized storage.
• Redis not used for config: From the deployment, it looks like Redis isn’t used for MCP server configs—just for some session or cache data.

Questions:

• Is there an up-to-date, comprehensive guide for deploying Langflow at scale in production on Kubernetes?
• How does the community recommend managing MCP server configs across multiple pods?
• Are there best practices or examples for leveraging Redis or other persistent backend stores for these configurations?
• Is there a way to deploy the backend/frontend independently without these issues?
• Would appreciate any guidance or pointers, especially from teams who have Langflow running reliably at scale.

Image used is v1.6.8 and also tried to build v1.7.0 from main branch.

Thanks!

[flefevre]

We’re running into many of the same challenges when trying to deploy Langflow at scale, particularly around exposing flows reliably as MCP servers across multiple replicas and managing their configurations.

Specifically:

We’ve also seen issues with flows/MCP server configurations not persisting across pods when running with the official Helm chart (e.g., ephemeral configs when pods restart or scale), which makes it hard to serve flows consistently.

There’s a strong need to expose flows as stable MCP tools, ideally with integration baked into the Helm chart or deployment manifests so that we don’t lose tool registrations on scaling. Current defaults appear geared more toward development than production use.

We also need OAuth-based authentication support for securing access to Langflow (both API/MCP), beyond simple API keys, to enable enterprise use cases and integrations with identity providers. Recent releases do mention OAuth support in documentation for MCP servers, but production-ready guidance/examples would help.

Has anyone successfully deployed a fully persistent and scalable Langflow + MCP setup (e.g., with centralized state storage, consistent MCP exposure, and OAuth authentication) in Kubernetes with Helm?