Feature

MCP Tools

Model Context Protocol for extending LLM capabilities with function calling and tool execution.

Overview

MCP (Model Context Protocol) enables LLMs to use external tools and functions. ZSE includes built-in tools and supports custom tool definitions compatible with OpenAI's function calling format.

Function Calling

Parse and execute tool calls from LLM output

Built-in Tools

Calculator, datetime, JSON parser, string ops

OpenAI Compatible

Same format as OpenAI function calling

JSON schema tool definitions
Automatic tool call parsing from LLM output
Built-in utility tools
Custom tool registration
OpenAI-compatible function calling format

Built-in Tools

ZSE includes several utility tools out of the box:

Tool	Description	Example Input
calculator	Math expressions (sqrt, sin, cos, log, etc.)	sqrt(16) + 2**3
datetime	Current date/time with timezone support	America/New_York
parse_json	Parse JSON and extract data	{"key": "value"}
string_ops	String operations (upper, lower, split, etc.)	upper: hello world

Calculator

bash

# Execute calculator tool
curl -X POST http://localhost:8000/api/tools/execute \
  -H "Content-Type: application/json" \
  -d '{
    "tool": "calculator",
    "input": "sqrt(144) + sin(3.14159/2)"
  }'
 
# Response:
# {"result": 13.0, "success": true}

Supported functions: sqrt, sin, cos, tan, log, log10, exp, abs, round, floor, ceil

Datetime

bash

# Get current time
curl -X POST http://localhost:8000/api/tools/execute \
  -H "Content-Type: application/json" \
  -d '{
    "tool": "datetime",
    "input": "UTC"
  }'
 
# Response:
# {
#   "result": {
#     "datetime": "2026-02-26T10:30:00+00:00",
#     "timezone": "UTC",
#     "unix_timestamp": 1772092200
#   },
#   "success": true
# }

Using Tools

List Available Tools

bash

# List all registered tools
curl http://localhost:8000/api/tools/
 
# Response:
# {
#   "tools": [
#     {
#       "name": "calculator",
#       "description": "Evaluate mathematical expressions",
#       "parameters": {...}
#     },
#     ...
#   ]
# }

Execute Tool

python

import requests
 
# Execute a tool directly
response = requests.post(
    "http://localhost:8000/api/tools/execute",
    json={
        "tool": "calculator",
        "input": "2 ** 10"
    }
)
print(response.json())
# {"result": 1024, "success": true}

Parse Tool Calls from Text

bash

# Parse tool calls from LLM output
curl -X POST http://localhost:8000/api/tools/parse \
  -H "Content-Type: application/json" \
  -d '{
    "text": "I need to calculate sqrt(256). Let me use the calculator tool."
  }'
 
# Response:
# {
#   "tool_calls": [
#     {"tool": "calculator", "input": "sqrt(256)"}
#   ]
# }

Parse and Execute

bash

# Parse and execute tool calls in one step
curl -X POST http://localhost:8000/api/tools/process \
  -H "Content-Type: application/json" \
  -d '{
    "text": "What time is it in Tokyo? Use datetime tool with Asia/Tokyo timezone."
  }'
 
# Response:
# {
#   "results": [
#     {
#       "tool": "datetime",
#       "input": "Asia/Tokyo",
#       "result": {"datetime": "2026-02-26T19:30:00+09:00", ...},
#       "success": true
#     }
#   ]
# }

Custom Tools

python

from zse.api.server.mcp import MCPRegistry
 
# Get the registry
registry = MCPRegistry()
 
# Define a custom tool
@registry.tool(
    name="weather",
    description="Get current weather for a location",
    parameters={
        "type": "object",
        "properties": {
            "location": {
                "type": "string",
                "description": "City name or coordinates"
            },
            "units": {
                "type": "string",
                "enum": ["celsius", "fahrenheit"],
                "default": "celsius"
            }
        },
        "required": ["location"]
    }
)
def get_weather(location: str, units: str = "celsius"):
    # Your implementation here
    return {"temperature": 22, "condition": "sunny", "units": units}
 
# Now the tool is available via API
# POST /api/tools/execute {"tool": "weather", "input": {"location": "Tokyo"}}

Custom tools are persisted in the server session. For permanent tools, add them to your server startup script.

OpenAI Format

Get tools in OpenAI-compatible function calling format:

Get Functions

bash

# Get tools in OpenAI format
curl http://localhost:8000/api/tools/openai/functions
 
# Response:
# {
#   "functions": [
#     {
#       "name": "calculator",
#       "description": "Evaluate mathematical expressions",
#       "parameters": {
#         "type": "object",
#         "properties": {
#           "expression": {
#             "type": "string",
#             "description": "Math expression to evaluate"
#           }
#         },
#         "required": ["expression"]
#       }
#     }
#   ]
# }

Chat with Tools

python

import requests
 
# Get available tools
tools_response = requests.get(
    "http://localhost:8000/api/tools/openai/functions"
)
tools = tools_response.json()["functions"]
 
# Chat with tool support
response = requests.post(
    "http://localhost:8000/v1/chat/completions",
    json={
        "model": "qwen-7b",
        "messages": [
            {"role": "user", "content": "What is 15% of 280?"}
        ],
        "tools": tools,
        "tool_choice": "auto"
    }
)
 
result = response.json()
# The model may return a tool call which you can execute
if result["choices"][0].get("tool_calls"):
    tool_call = result["choices"][0]["tool_calls"][0]
    # Execute the tool call...

API Reference

Endpoint	Method	Description
/api/tools/	GET	List all available tools
/api/tools/execute	POST	Execute a specific tool
/api/tools/parse	POST	Parse tool calls from text
/api/tools/process	POST	Parse and execute tool calls
/api/tools/openai/functions	GET	Get OpenAI-compatible format

← RAG Module

CLI Commands →