Esta es una demostración mínima de MCP escrita a mano sin el SDK de Python de MCP. Contiene el flujo completo:
nanomcp.server actúa como servidor MCP, enviando y recibiendo JSON-RPC a través de stdio.
nanomcp.cli actúa como cliente/host MCP, inicia el servidor y realiza initialize, tools/list y tools/call.
El comando chat invoca las Chat Completions de OpenAI. Después de que el modelo devuelve una llamada a función/herramienta, la CLI la convierte en una tools/call de MCP y luego envía el resultado de la herramienta de vuelta al modelo para generar la respuesta final.
En resumen: function call es la capacidad de la API del modelo para que "el modelo le diga a tu aplicación qué función quiere llamar"; MCP es el protocolo de conexión para que "tu aplicación descubra y llame a herramientas externas/servicios de contexto mediante un protocolo unificado".
Más específicamente:
Function/tool calling ocurre entre LLM API <-> tu aplicación. El modelo no ejecuta realmente la función, solo devuelve una intención de llamada como {"name":"get_weather","arguments":{...}}.
MCP ocurre entre tu aplicación <-> servidor MCP. El servidor MCP expone una lista de herramientas y puntos de entrada de ejecución, como tools/list y tools/call.
El host/cliente es el intermediario. Primero obtiene el esquema de herramientas del servidor MCP y convierte estos esquemas en herramientas de la API del modelo; después de que el modelo selecciona una herramienta, el host/cliente llama al servidor MCP.
El flujo en este proyecto es:
用户问题
-> nanomcp.cli
-> OpenAI Chat Completions tools=function schemas
<- 模型返回 tool_calls
-> nanomcp.cli 把 tool_call 映射为 MCP tools/call
-> nanomcp.server 执行 get_weather 或 find_files
<- MCP tool result
-> nanomcp.cli 把结果发回模型
<- 模型最终回答Por lo tanto, no están en el mismo nivel:
Function call: 模型 API 的工具选择/参数生成机制
MCP: 应用连接工具服务器的标准协议Related MCP server: MCP Server Demo
nanomcp/
nanomcp/
cli.py # MCP client + model caller
server.py # hand-written MCP server over stdio
tests/
test_protocol.py
pyproject.toml
README.mdEjecuta en el directorio del proyecto:
cd ~/Desktop/nanomcp
python3 -m nanomcp.cli list-toolsLlamar directamente a la herramienta de clima:
python3 -m nanomcp.cli call get_weather '{"location":"Shanghai","unit":"celsius"}'Llamar directamente a la herramienta de fecha y hora actual:
python3 -m nanomcp.cli call get_current_datetime '{"timezone":"Asia/Shanghai"}'Llamar directamente a la herramienta de búsqueda de archivos:
python3 -m nanomcp.cli call find_files '{"query":"*.pdf","max_results":5}'Por defecto, solo busca en ~/Desktop. Puedes ampliar o reducir temporalmente el directorio raíz de búsqueda:
NANOMCP_FILE_ROOT=~/Desktop/nanomcp python3 -m nanomcp.cli call find_files '{"query":"*.py"}'Se requiere una clave de API de OpenAI. Aquí no se utiliza el SDK de Python de OpenAI, sino que se envía HTTP directamente mediante la biblioteca estándar urllib.
Se recomienda escribir la configuración local en .env:
cd ~/Desktop/nanomcp
cp .envtemplate .envLuego edita .env:
OPENAI_API_KEY=你的 key
OPENAI_BASE_URL=https://api.openai.com/v1
NANOMCP_MODEL=gpt-4.1-mini
NANOMCP_TIMEZONE=Asia/Shanghai.env será leído automáticamente por la CLI y ya ha sido ignorado por .gitignore.
cd ~/Desktop/nanomcp
python3 -m nanomcp.cli chat "上海今天天气怎么样？顺便帮我找桌面上的 PDF 文件"El modelo por defecto es gpt-4.1-mini. Puedes cambiarlo:
NANOMCP_MODEL=gpt-5-mini python3 -m nanomcp.cli chat "找一下这个项目里的 py 文件"Si utilizas una pasarela compatible con OpenAI:
OPENAI_BASE_URL=http://localhost:8000/v1 python3 -m nanomcp.cli chat "上海天气怎么样？"Verificación ligera de la configuración local y el servidor MCP:
python3 -m nanomcp.cli doctorSi chat muestra OpenAI API quota is exhausted (429 insufficient_quota), significa que la API del modelo rechazó la solicitud: el proyecto al que pertenece la OPENAI_API_KEY actual no tiene cuota disponible o la facturación no está activada. Esto no es un fallo del servidor MCP, ya que la solicitud fue rechazada antes de que el modelo devolviera la llamada a la herramienta.
Orden de verificación:
python3 -m nanomcp.cli doctor
echo "$OPENAI_API_KEY"
cat .env
python3 -m nanomcp.cli call get_weather '{"location":"Shanghai"}'
OPENAI_BASE_URL=http://localhost:8000/v1 python3 -m nanomcp.cli chat "上海天气怎么样？"El primer comando muestra de forma segura la configuración válida, si el shell sobrescribe .env y si el servidor MCP puede listar las herramientas.
El segundo comando confirma si la clave ya está configurada en el shell.
El tercer comando confirma la configuración local en .env.
El cuarto comando verifica si el flujo MCP local es normal, sin depender de la API del modelo.
El quinto comando demuestra cómo cambiar a una pasarela compatible con OpenAI.
Si aún utilizas la API oficial de OpenAI, necesitas cambiar a una clave/proyecto con cuota, o verificar la facturación y los permisos del modelo.
El clima por defecto es un dato de demostración determinista, conveniente para aprender el flujo del protocolo sin red o claves de terceros. Para probar una consulta real:
NANOMCP_LIVE_WEATHER=1 python3 -m nanomcp.cli call get_weather '{"location":"Shanghai"}'El clima real utiliza https://wttr.in, y volverá automáticamente a los datos de demostración en caso de fallo.
cd ~/Desktop/nanomcp
python3 -m unittest discover -s testsCobertura de pruebas:
MCP initialize
MCP tools/list
MCP tools/call get_weather
MCP tools/call find_files
MCP tools/call get_current_datetime
Observa openai_tools_from_mcp() en nanomcp/cli.py: convierte el esquema de herramientas de MCP al esquema de funciones de OpenAI.
Observa run_chat(): llama a mcp.call_tool() después de recibir tool_calls del modelo. Este es el punto de conexión entre MCP y function call.
Observa main() en nanomcp/server.py: solo lee stdin y escribe en stdout, cada línea es un JSON-RPC. El servidor no conoce OpenAI ni interactúa directamente con el modelo.
Find local files by name under the allowed root. The default root is ~/Desktop. Set NANOMCP_FILE_ROOT to change it.
| Name | Required | Description | Default |
|---|---|---|---|
| query | Yes | Filename substring or glob pattern, such as *.pdf. | |
| root | No | Optional subdirectory under NANOMCP_FILE_ROOT. | |
| max_results | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description carries full burden. It mentions the root directory and default, but it does not disclose important behaviors such as case sensitivity, recursion depth, glob pattern handling, permissions, or the structure of returned results.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is extremely concise: two sentences. The first sentence states purpose and scope, the second provides configuration info. Every sentence adds value with no redundancy.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given that there is no output schema, the description should hint at return format or behavior. It does not mention what is returned (file paths, metadata), sorting, recursion, or error handling. The tool is simple but the agent may need more context for correct invocation.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The input schema already describes two of three parameters (query and root). The description adds context about the root default and environment variable configuration, but does not enhance understanding of max_results or clarify glob pattern syntax beyond what the schema provides.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool's function: 'Find local files by name under the allowed root.' It specifies the scope (local files) and the constraint (under a root). The siblings are unrelated (datetime and weather), so there is no ambiguity.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides no explicit guidance on when to use this tool versus alternatives. It does not mention when not to use it or any prerequisites. Given that siblings are unrelated, implicit guidance is minimal.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Get the current date, time, and weekday. Use this for questions about today, current time, current date, or weekday.
| Name | Required | Description | Default |
|---|---|---|---|
| timezone | No | IANA timezone name, such as Asia/Shanghai or America/New_York. Defaults to NANOMCP_TIMEZONE or Asia/Shanghai. | Asia/Shanghai |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations exist, so the description carries full burden. It adequately describes the output (date, time, weekday) and timezone parameter. However, it does not mention that the operation is read-only, instantaneous, or any potential dependencies, leaving some behavioral details implicit.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is two sentences, concise and front-loaded with the core function. Every sentence serves a purpose without redundancy.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the tool's simplicity (one optional parameter, no output schema), the description fully covers what the tool does, its possible output, and appropriate use cases. No gaps remain.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
With full schema coverage (100%), the description adds no new parameter details beyond the schema. The schema already describes the timezone parameter well, so the description provides minimal added value, meeting the baseline of 3.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool retrieves current date, time, and weekday. It explicitly lists use cases like 'today, current time, current date, or weekday', and siblings are unrelated, making purpose unambiguous.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description directly states when to use the tool ('for questions about today, current time, current date, or weekday'). It does not provide exclusions or alternatives, but given the simplicity and distinct siblings, this is sufficient.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Get current weather for a city. By default this returns deterministic demo data. Set NANOMCP_LIVE_WEATHER=1 to try wttr.in.
| Name | Required | Description | Default |
|---|---|---|---|
| location | Yes | City or place name, for example Shanghai. | |
| unit | No | Temperature unit. | celsius |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations exist, so the description carries the burden. It discloses the demo/live behavior and environment variable, but lacks details on return format, error handling, or external API dependencies.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two sentences efficiently define purpose and critical behavioral context. No superfluous text.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple weather tool, the description covers core purpose and key behavioral nuance. However, it omits return value structure or typical properties, which would help the agent understand the output.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Input schema covers 100% of parameters with descriptions. The description adds no additional parameter meaning beyond what the schema provides, meeting baseline.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states 'Get current weather for a city,' using a specific verb and resource, and distinguishes from siblings like find_files and get_current_datetime.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
It explains the default demo mode and how to switch to live data, providing context for when to expect real or synthetic data. No explicit alternatives or exclusions but sufficient for this tool.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v0.1.0find_filesget_current_datetimeget_weatherScored across 3 tools
Each tool has a clear, distinct purpose: file search, datetime, and weather. No overlap or ambiguity.
All tool names follow the verb_noun snake_case pattern consistently: find_files, get_current_datetime, get_weather.
Three tools is small but appropriate for a 'nano' server intended as a minimal utility collection. Not too few given its scope.
The tools cover only three disparate areas with no clear domain. As a general utility set, common operations like calculations or text processing are missing, but it may be intentionally limited.
Nifty's MCP server — exposes tasks, projects, messages, and files as tools for AI agents.
A comprehensive Model Context Protocol (MCP) server that enables AI assistants to interact with yo…
