Flama

Building an MCP Server with Flama

How to build and serve a Model Context Protocol server that exposes tools, resources, and prompts to AI agents

Publication
Reading Time
~ 12 min read

Building an MCP Server with Flama

Serving a model is only half the story. The other half is giving AI agents access to your world: the functions they can call, the data they can read, and the prompt templates they can reuse. The Model Context Protocol (MCP) is the open standard for exactly that, and Flama provides native, first-class support for building MCP servers with nothing more than a few decorators on plain Python functions.

In this post, we walk through building a complete MCP server with Flama. We will expose tools, resources, and prompts to any MCP-capable client, and we will explore the advanced extensions for background tasks, interactive input, and embedded user interfaces. By the end, you will have a running server that any AI assistant can discover and call.

Before we dive into the details, we recommend you to have the following resources at hand:

Table of contents

What is MCP?

The Model Context Protocol is an open standard that lets AI applications connect to external capabilities through a uniform interface. An MCP server advertises three kinds of capability:

  • Tools: functions the model can invoke.
  • Resources: data the model can read, addressed by URI.
  • Prompts: reusable prompt templates with arguments.

Clients (AI assistants, agent frameworks, IDEs) discover these capabilities and call them over JSON-RPC, a lightweight remote-procedure-call protocol that exchanges JSON messages.

Flama implements the stateless 2026-07-28 revision of the protocol. Rather than negotiating a session through an initialize handshake, every request is self-contained, carrying its protocol version and capabilities in a _meta object and its routing data in Mcp-Method / Mcp-Name headers. This makes MCP servers trivial to scale horizontally, since no per-client state is held between calls.

Why does this matter?

  • Interoperability: Any MCP-capable client can use your tools without bespoke integration code.
  • Reuse: The same Python functions that power your API can be exposed to AI agents with a single decorator.
  • Type safety: Flama derives each tool's input and output JSON Schema from the handler's type hints, so clients receive accurate, self-contained contracts.

Setting up the project

All examples in this post assume Flama has been installed with the pydantic extras via uv:

uv pip install "flama[pydantic]"

Alternatively, you can run any command without a prior install by using uvx --from "flama[pydantic]" flama ..., but for brevity we assume Flama is already installed throughout.

Registering an MCP server

An MCP server in Flama is a named registry that you mount on your application at a specific URL path. The add_server method both creates the server and mounts it, so a single application can host several independent servers:

import flamafrom flama import Flama
app = Flama(    openapi={        "info": {            "title": "MCP Server API",            "version": "1.0.0",            "description": "A Model Context Protocol server built with Flama 🔥",        },    },)
app.mcp.add_server("/mcp/tools/", "tools", version="2.0.0", instructions="Flama demo MCP tools server")

This registers a server named tools, reachable at /mcp/tools/. The version parameter declares the server's semantic version, and instructions provides a human-readable description that clients can display. With the server in place, you populate it by name: every tool, resource, and prompt decorator takes an mcp argument identifying which server the capability belongs to.

Exposing tools

A tool is a function the model can invoke. Declare one with the tool decorator, pointing it at the target server through the mcp argument. Flama infers the tool's input and output schema from the handler's type hints:

@app.mcp.tool("add", description="Add two integers", mcp="tools")def add(a: int, b: int) -> int:    return a + b

Tools may be synchronous or asynchronous. When you omit the name, the function's own name is used; when you omit the description, its docstring is used instead. The parameters and return annotation become the tool's inputSchema and outputSchema, advertised to clients verbatim.

Here is an asynchronous tool that returns a string:

@app.mcp.tool("greet", description="Greet someone by name", mcp="tools")async def greet(name: str) -> str:    return f"Hello, {name}!"

Let us verify the tool works. Start the application:

flama run app:app

And call it with curl:

curl -s -X POST http://127.0.0.1:8000/mcp/tools/ \  -H 'Content-Type: application/json' \  -H 'Mcp-Method: tools/call' \  -H 'Mcp-Name: add' \  -H 'MCP-Protocol-Version: 2026-07-28' \  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"add","arguments":{"a":2,"b":3}}}'

The server responds with a JSON-RPC result:

{  "jsonrpc": "2.0",  "id": 1,  "result": {    "content": [{"type": "text", "text": "5"}],    "structuredContent": 5  }}

The structuredContent field carries the typed return value, while content provides a text representation for clients that prefer unstructured output.

Exposing resources

A resource is readable data addressed by a URI. The resource decorator registers one on the named server:

import json
@app.mcp.resource("config://app", name="config", description="Application configuration",                  mime_type="application/json", mcp="tools")def config():    return json.dumps({"debug": True, "name": "flama-mcp"})

Resources are listed and read by their URI, so a client fetches the configuration above by requesting config://app. The MIME type tells the client how to interpret the content.

To read the resource:

curl -s -X POST http://127.0.0.1:8000/mcp/tools/ \  -H 'Content-Type: application/json' \  -H 'Mcp-Method: resources/read' \  -H 'Mcp-Name: config://app' \  -H 'MCP-Protocol-Version: 2026-07-28' \  -d '{"jsonrpc":"2.0","id":1,"method":"resources/read","params":{"uri":"config://app"}}'
{  "jsonrpc": "2.0",  "id": 1,  "result": {    "contents": [      {        "uri": "config://app",        "mimeType": "application/json",        "text": "{\"debug\": true, \"name\": \"flama-mcp\"}"      }    ]  }}

Exposing prompts

A prompt is a named, reusable prompt template. The prompt decorator registers one on the named server, deriving its arguments from the handler's parameters:

@app.mcp.prompt("summarise", description="Summarise the given text", mcp="tools")def summarise(text: str):    return f"Summarise the following:\n\n{text}"

Prompts are listed by name and rendered with arguments supplied by the client. Here text becomes the single required argument. To get the rendered prompt:

curl -s -X POST http://127.0.0.1:8000/mcp/tools/ \  -H 'Content-Type: application/json' \  -H 'Mcp-Method: prompts/get' \  -H 'Mcp-Name: summarise' \  -H 'MCP-Protocol-Version: 2026-07-28' \  -d '{"jsonrpc":"2.0","id":1,"method":"prompts/get","params":{"name":"summarise","arguments":{"text":"Flama is great"}}}'
{  "jsonrpc": "2.0",  "id": 1,  "result": {    "messages": [      {        "role": "user",        "content": {"type": "text", "text": "Summarise the following:\n\nFlama is great"}      }    ]  }}

Advanced extensions

The 2026-07-28 protocol defines optional extensions, all supported natively by Flama. A server advertises the extensions it uses in its discovery capabilities, so clients negotiate them per request.

Background tasks

Long-running tools can run as background Tasks rather than blocking the call. Pass task=True and the server returns a task handle the client can poll:

@app.mcp.tool("square", task=True, description="Square a number as a background task", mcp="tools")async def square(x: int) -> int:    return x * x

When a client calls square, the server may return the result directly for fast operations, or issue a task token for truly long-running computations that the client can poll until completion.

Elicitation

A tool can pause mid-call to elicit further input from the user. The handler declares a parameter annotated with Elicitation to read the answers gathered so far, and returns Elicit.require(...) to request more:

from flama.mcp.data_structures import Elicit, Elicitation
@app.mcp.tool("confirm", description="Confirm an action through an elicitation round-trip", mcp="tools")def confirm(elicitation: Elicitation) -> str:    if "confirm" not in elicitation:        return Elicit.require("Are you sure?", {"type": "boolean"}, name="confirm")    return f"confirmed={elicitation['confirm']}"

The elicitation parameter is supplied by the server and excluded from the tool's input schema, so it never appears as a tool argument the client must fill. Because the protocol is stateless, the answers gathered so far are round-tripped through an opaque continuation token the client echoes back on the retry.

When the client calls confirm without prior answers, it receives a response with resultType: "inputRequired" and a schema describing what the server needs. The client collects that input from the user and retries, this time carrying the gathered answers.

MCP Apps

A tool can declare a prefetchable user-interface template (an MCP App) that hosts render alongside its result. Register the template with app_template and point the tool at it with ui_template:

@app.mcp.app_template("ui://widget", name="widget", description="A small UI widget", mcp="tools")def widget():    return "<html><body><h1>Flama widget</h1></body></html>"
@app.mcp.tool("with_ui", description="A tool that declares a prefetchable UI template",              ui_template="ui://widget", mcp="tools")def with_ui() -> str:    return "rendered"

Clients that support MCP Apps can prefetch the template and render it alongside the tool's result, providing a richer interactive experience.

Multiple servers in one application

A single Flama application can host as many MCP servers as you need, each under its own path. This is useful for separating concerns or versioning different sets of capabilities:

app.mcp.add_server("/mcp/tools/", "tools", version="2.0.0", instructions="Flama demo MCP tools server")app.mcp.add_server("/mcp/math/", "math", version="2.0.0")

Each server is independent. Tools, resources, and prompts are bound to their server by the mcp argument:

@app.mcp.tool("multiply", description="Multiply two integers", mcp="math")def multiply(a: int, b: int) -> int:    return a * b

A tools/list request to /mcp/tools/ returns only the tools registered on the tools server, while a request to /mcp/math/ returns only multiply. Clients discover each server independently.

The complete application

Putting it all together, here is the full application. It registers two MCP servers on a single Flama app, populates them with tools (sync, async, background task, elicitation, UI template), a resource, and a prompt:

import json
import flamafrom flama import Flamafrom flama.mcp.data_structures import Elicit, Elicitation
app = Flama(    openapi={        "info": {            "title": "MCP Server API",            "version": "1.0.0",            "description": "A Model Context Protocol server built with Flama 🔥",        },    },)
app.mcp.add_server("/mcp/tools/", "tools", version="2.0.0", instructions="Flama demo MCP tools server")app.mcp.add_server("/mcp/math/", "math", version="2.0.0")
@app.mcp.tool("add", description="Add two integers", mcp="tools")def add(a: int, b: int) -> int:    return a + b
@app.mcp.tool("greet", description="Greet someone by name", mcp="tools")async def greet(name: str) -> str:    return f"Hello, {name}!"
@app.mcp.tool("square", task=True, description="Square a number as a background task", mcp="tools")async def square(x: int) -> int:    return x * x
@app.mcp.tool("confirm", description="Confirm an action through an elicitation round-trip", mcp="tools")def confirm(elicitation: Elicitation) -> str:    if "confirm" not in elicitation:        return Elicit.require("Are you sure?", {"type": "boolean"}, name="confirm")    return f"confirmed={elicitation['confirm']}"
@app.mcp.resource("config://app", name="config", description="Application configuration",                  mime_type="application/json", mcp="tools")def config():    return json.dumps({"debug": True, "name": "flama-mcp"})
@app.mcp.prompt("summarise", description="Summarise the given text", mcp="tools")def summarise(text: str):    return f"Summarise the following:\n\n{text}"
@app.mcp.app_template("ui://widget", name="widget", description="A small UI widget", mcp="tools")def widget():    return "<html><body><h1>Flama widget</h1></body></html>"
@app.mcp.tool("with_ui", description="A tool that declares a prefetchable UI template",              ui_template="ui://widget", mcp="tools")def with_ui() -> str:    return "rendered"
@app.mcp.tool("multiply", description="Multiply two integers", mcp="math")def multiply(a: int, b: int) -> int:    return a * b
if __name__ == "__main__":    flama.run(flama_app=app, server_host="0.0.0.0", server_port=8000)

Save this as app.py and run it:

python app.py

The server starts on port 8000 with both MCP endpoints ready.

Testing with curl

Once the application is running, you can exercise every capability from the command line.

List available tools on the tools server:

curl -s -X POST http://127.0.0.1:8000/mcp/tools/ \  -H 'Content-Type: application/json' \  -H 'Mcp-Method: tools/list' \  -H 'MCP-Protocol-Version: 2026-07-28' \  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

The response lists five tools (add, confirm, greet, square, with_ui), each with its full input and output schema derived from the Python type hints.

Call a tool on the math server:

curl -s -X POST http://127.0.0.1:8000/mcp/math/ \  -H 'Content-Type: application/json' \  -H 'Mcp-Method: tools/call' \  -H 'Mcp-Name: multiply' \  -H 'MCP-Protocol-Version: 2026-07-28' \  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"multiply","arguments":{"a":4,"b":5}}}'
{  "jsonrpc": "2.0",  "id": 1,  "result": {    "content": [{"type": "text", "text": "20"}],    "structuredContent": 20  }}

Read a resource:

curl -s -X POST http://127.0.0.1:8000/mcp/tools/ \  -H 'Content-Type: application/json' \  -H 'Mcp-Method: resources/read' \  -H 'Mcp-Name: config://app' \  -H 'MCP-Protocol-Version: 2026-07-28' \  -d '{"jsonrpc":"2.0","id":1,"method":"resources/read","params":{"uri":"config://app"}}'

Get a rendered prompt:

curl -s -X POST http://127.0.0.1:8000/mcp/tools/ \  -H 'Content-Type: application/json' \  -H 'Mcp-Method: prompts/get' \  -H 'Mcp-Name: summarise' \  -H 'MCP-Protocol-Version: 2026-07-28' \  -d '{"jsonrpc":"2.0","id":1,"method":"prompts/get","params":{"name":"summarise","arguments":{"text":"Flama is great"}}}'

Every request follows the same pattern: a POST to the server's path, with Mcp-Method identifying the operation, Mcp-Name identifying the target, and MCP-Protocol-Version declaring the protocol revision.

Conclusions

Flama makes the journey from "I have Python functions" to "AI agents can discover and call them" as short as possible. The MCP support requires no configuration files, no code generation, and no external tooling. You write plain Python functions, decorate them, and the framework handles the rest:

  1. add_server: Mount a named MCP server at any path.
  2. @tool: Expose a function as an invocable tool with full schema inference.
  3. @resource: Expose data at a URI for clients to read.
  4. @prompt: Expose a reusable prompt template with typed arguments.
  5. Extensions: Background tasks, elicitation, and MCP Apps for richer interactions.

Because the protocol is stateless, your servers scale horizontally without sticky sessions. Because the schema is derived from type hints, clients receive accurate contracts without manual specification. And because multiple servers can live in a single application, you can organise capabilities by domain, version, or access level.

In upcoming posts, we will explore how to combine MCP servers with LLM serving to build fully autonomous agent architectures where the model and its tools live in the same application.

References

Support our work

If you find Flama useful for building robust Machine Learning and Generative AI APIs, we'd be thrilled if you showed your support by giving us a ⭐ on GitHub. Your stars are the best fuel for our development efforts!

You can also stay updated with the latest news and development threads by following us on 𝕏.

About the authors

Vortico: We specialize in software development, helping businesses enhance and expand their AI and technology capabilities.