Integrating FastMCP in a Secure, Multi-Tenant, Production-Grade AI System on AWS EKS

[awsplato]

Hi FastMCP Team & Community,

We're in the process of integrating FastMCP into a production-grade AI application for an enterprise client, and we're evaluating the best approach to do so securely and scalably in a multi-tenant environment.

---

Context

• Platform: AWS EKS
• Architecture: We have a central gateway inside the EKS cluster that handles AuthN/AuthZ, validates tokens, and routes requests to backend services (including future MCP services).
• MCP Exposure: MCP endpoints will be behind the gateway, with no direct public access.

---

Our Use Case with FastMCP

We're considering using FastMCP to manage and invoke MCP servers (e.g., Microsoft-provided tools). However, we face a few architectural requirements and constraints:

---

❓ Key Questions & Needs

1. Secret Management Without Env Variables

Some MCP servers (like Microsoft’s) rely on environment variables for credentials and config.

In our setup:

• We use AWS Secrets Manager to manage secrets per tenant.
• We prefer not to pass secrets via environment variables for security and scalability.
• Instead, we plan to inject secrets at runtime from Secrets Manager.

Q: Can FastMCP be extended or configured to support secret retrieval dynamically (e.g., via a wrapper or plugin) before invoking the MCP server?

---

2. Custom Access Control per Tool

Before invoking any MCP tool, we need to validate:

• Whether the authenticated user has access to a specific tool (fine-grained AuthZ).
• Possibly inject tenant-specific configurations or secrets.

Q: Is there a recommended way to introduce custom logic/hooks in FastMCP for access checks and request pre-processing before the MCP server is called?

---

3. Best Practices for Multi-Tenant Systems

• We're looking to avoid duplicating logic for every tenant.
• We need centralized but configurable control over secrets, permissions, and routing.

Q: Has anyone in the community implemented FastMCP in a multi-tenant setup? Any best practices or architectural guidance?

---

What We're Considering

We may end up:

• Creating a wrapper service that:
  • Fetches secrets from Secrets Manager.
  • Performs AuthZ checks.
  • Then delegates to FastMCP to invoke the target MCP server.

But we’d love to avoid duplicating logic if FastMCP supports (or could support) such extensibility natively.

---

🙏 Ask to the Community

We’d really appreciate any insights, advice, or examples from others who have used FastMCP in secure, multi-tenant environments.
We’re happy to contribute back anything we build if it’s useful to the broader community.

Thanks in advance!

[awsplato]

@jlowin ++

[jeremie-cisco]

Hi @awsplato! Came here to say that we are also interested in a multi-tenant setup of FastMCP. Have you made any progress in your research since you posted here?

For context, my team is building an MCP functionality for our product which has thousands of customers, each with their own Authorization Server. Deploying a set of single tenant MCP servers for that many customers is something we'd want to avoid, especially given how much traffic these web servers are able to handle. Instead, we'd like to at least make the Auth piece support dynamic tenant configuration, and ideally have session management implement multi-tenant best practices. We're happy to contribute to that effort as well.