What is MCP?

The Model Context Protocol (MCP) is an open standard for connecting AI assistants to external tools and data sources. Instead of hardcoding integrations, MCP provides a clean interface where servers expose tools, resources, and prompts that any MCP client can discover and use.

The problem: deploying MCP servers is still manual

Most MCP server implementations are tightly coupled to a specific set of tools. You write Python functions, wire them into a server, build a Docker image, and deploy. When you want to add a new tool, you rebuild and redeploy.

This works for small setups, but it does not scale when:

Different teams need different tool sets on the same server
Knowledge bases change frequently and live in S3 or Git
You want to try a new tool without rebuilding the image
You need the same server in Docker Compose, Kubernetes, and AWS ECS

FastMCP Server: dynamic tool loading, zero rebuilds

FastMCP Server is a standalone Docker image that separates the MCP runtime from the tools. You deploy the server once and load tools dynamically from three sources:

Inline mounts — volume-mount Python files directly (highest precedence)
S3-compatible storage — AWS S3, MinIO, Cloudflare R2 (medium precedence)
Git repositories — clone tools from any Git repo at startup (lowest precedence)

The merge precedence means you can have a shared tool library in S3, override specific tools with inline mounts for testing, and keep everything in Git for version control.

Quick example: Docker

# Mount local tools and start the server
docker run -d -p 8000:8000 \
  -v ./my-tools:/app/inline/tools \
  -v ./my-knowledge:/app/inline/knowledge \
  docker.io/helmforge/fastmcp-server:0.2.0

Quick example: Kubernetes (Helm)

helm install fastmcp-server oci://ghcr.io/helmforgedev/helm/fastmcp-server \
  --set sources.s3.enabled=true \
  --set sources.s3.bucket=mcp-tools \
  --set auth.type=bearer \
  --set auth.bearer.token=my-token

What you can load

Type	Format	Example
Tools	Python functions	`def greet(name: str) -> str`
Resources	Python with `RESOURCE_URI`	Server status, API endpoints
Prompts	Python functions returning strings	Summarization templates
Knowledge	Markdown files	Architecture docs, runbooks, FAQs

Knowledge files are automatically served as MCP resources at knowledge://{filename} URIs, making them available for RAG and context injection by AI assistants.

Authentication built-in

FastMCP Server supports two authentication modes without any reverse proxy:

Bearer token — simple static token via MCP_AUTH_TYPE=bearer
JWT — full OIDC/JWT verification with JWKS endpoint via MCP_AUTH_TYPE=jwt

Runs anywhere

The image is not tied to Kubernetes. It works with:

docker run — local development
Docker Compose — multi-service setups with MinIO
Docker Swarm — production without Kubernetes
AWS ECS / Fargate — serverless containers
Kubernetes via the HelmForge Helm chart — full production with ingress, PVC, NetworkPolicy

Try it

docker run -d -p 8000:8000 docker.io/helmforge/fastmcp-server:0.2.0

The server starts with no tools (that is fine) and the healthcheck endpoint responds at http://localhost:8000/mcp. Mount your tools directory and they appear instantly.

Deploy MCP Servers Anywhere with FastMCP Server