FastMCP Server Development

FastMCP Server Development

FastMCP is a Python framework for building Model Context Protocol (MCP) servers. It provides a decorator-based API for exposing tools, resources, and prompts to AI assistants. This skill covers the core API for building production-ready MCP servers.

Server Initialization

from fastmcp import FastMCP

mcp = FastMCP(
    name="MyServer",
    instructions="A helpful server for data processing",
    version="1.0.0",
)

• name — display name for the server.
• instructions — system-level guidance for AI assistants connecting to this server.
• version — semantic version string.

Tools

Tools are functions that AI assistants can invoke. Decorate with @mcp.tool.

Basic tool

@mcp.tool
def add(a: int, b: int) -> int:
    """Add two numbers together."""
    return a + b

FastMCP auto-generates the JSON schema from the function signature and docstring. The docstring becomes the tool description shown to the AI.

Async tool

@mcp.tool
async def fetch_data(url: str) -> dict:
    """Fetch JSON data from a URL."""
    async with httpx.AsyncClient() as client:
        response = await client.get(url)
        return response.json()

Custom name and description

@mcp.tool(name="multiply", description="Multiply two numbers")
async def multiply_numbers(x: float, y: float) -> float:
    return x * y

Structured input with Pydantic

from pydantic import BaseModel

class SearchQuery(BaseModel):
    query: str
    max_results: int = 10
    include_metadata: bool = False

@mcp.tool
def search(params: SearchQuery) -> list[dict]:
    """Search with structured parameters."""
    return [{"result": f"Found for: {params.query}"}]

Use Pydantic models when a tool has more than 3-4 parameters or when validation is needed.

Context object

The Context parameter provides logging, progress reporting, and access to lifespan resources:

from fastmcp.server.context import Context

@mcp.tool
async def process_files(file_paths: list[str], ctx: Context) -> dict:
    """Process multiple files with progress reporting."""
    total = len(file_paths)
    await ctx.info(f"Starting to process {total} files")

    for i, path in enumerate(file_paths):
        await ctx.report_progress(i, total, f"Processing {path}")
        try:
            # process file...
            pass
        except Exception as e:
            await ctx.error(f"Failed to process {path}: {e}")

    await ctx.report_progress(total, total, "Complete")
    return {"processed": total}

Context methods:

• ctx.info(msg), ctx.debug(msg), ctx.warning(msg), ctx.error(msg) — structured logging.
• ctx.report_progress(current, total, message) — progress updates.
• ctx.lifespan_context — dict of resources from the lifespan handler.

Add ctx: Context as any parameter — FastMCP injects it automatically (it is not exposed to the AI).

Resources

Resources expose data that AI assistants can read. Decorate with @mcp.resource.

Static resource

@mcp.resource("config://settings")
def get_settings() -> str:
    """Return application settings."""
    return "debug=true\nlog_level=INFO"

JSON resource

@mcp.resource("data://report", mime_type="application/json")
def get_report() -> dict:
    """Return JSON data (automatically serialized)."""
    return {"users": 150, "active": 42}

Dynamic resource template

Use {parameter} in the URI for parameterized resources:

@mcp.resource("users://{user_id}/profile")
async def get_user_profile(user_id: str) -> dict:
    """Fetch user profile by ID."""
    return {"id": user_id, "name": f"User {user_id}"}

The AI sees a URI template and can fill in the parameter to read specific resources.

Prompts

Prompts are reusable templates that guide AI interactions. Decorate with @mcp.prompt.

Simple prompt

@mcp.prompt
def greeting() -> str:
    """A friendly greeting prompt."""
    return "Hello! How can I help you today?"

Parameterized prompt

@mcp.prompt
def analyze_code(language: str, code: str) -> str:
    """Generate a code analysis prompt."""
    return f"Please analyze this {language} code and suggest improvements:\n\n```{language}\n{code}\n```"

Running the Server

STDIO transport (default)

if __name__ == "__main__":
    mcp.run()

STDIO is the default transport. Use for local integrations (e.g., Claude Desktop).

HTTP transport (Streamable HTTP)

if __name__ == "__main__":
    mcp.run(transport="http", host="0.0.0.0", port=8000)

Use HTTP for remote/deployed servers. Clients connect to http://host:port/mcp.

Server Composition

Mount sub-servers to organize a large server into logical modules:

math_server = FastMCP("MathServer")
text_server = FastMCP("TextServer")

@math_server.tool
def add(a: int, b: int) -> int:
    return a + b

@text_server.tool
def uppercase(text: str) -> str:
    return text.upper()

# Main server mounts sub-servers
main = FastMCP("MainServer")
main.mount(math_server, namespace="math")   # tools: math_add
main.mount(text_server)                      # tools: uppercase (no prefix)

• namespace — optional prefix added to all tool/resource names from the sub-server.
• Keeps each module independently testable.

Common Patterns Quick Reference

Key Rules

1. Always type-annotate tool parameters and return types — FastMCP generates the schema from them.
2. Write clear docstrings — the docstring becomes the tool/resource/prompt description visible to the AI.
3. Use ctx: Context for logging and progress — do not use print().
4. Prefer Pydantic models for tools with 4+ parameters or when input validation is needed.
5. Use namespace when mounting sub-servers to avoid tool name collisions.
6. Choose the right transport — STDIO for local, HTTP for remote/deployed servers.

Additional Resources

Reference Files

For detailed API documentation and advanced patterns, consult:

• references/api-reference.md — Tool decorator options, resource MIME types, prompt patterns, Context API, authentication (JWT/OAuth2), server configuration
• references/patterns.md — Server composition, lifespan management, OpenAPI/FastAPI import, proxy servers, dependencies, error handling, deployment