Physics MCP Server

# Natural Language Interface Workflow Demonstrate end-to-end natural language to tool call conversion using the NLI system. ## Overview The Natural Language Interface (NLI) converts plain English physics requests into structured MCP tool calls, making complex calculations accessible through conversational interaction. ## Setup Ensure you have a local LM server running (e.g., LM Studio) for optimal performance: ```bash # Set environment variables export LM_BASE_URL="http://localhost:1234/v1" export DEFAULT_MODEL="llama-3.2-3b-instruct" # LM_API_KEY is optional for local servers ``` ## Example Workflow ### 1. Simple Calculation **Natural Language Input:** > "Calculate the derivative of x squared plus 3x plus 1" **NLI Processing:** ```json { "jsonrpc": "2.0", "id": "1", "method": "nli_parse", "params": { "text": "Calculate the derivative of x squared plus 3x plus 1" } } ``` **Expected NLI Output:** ```json { "tool_call": { "method": "cas", "params": { "action": "diff", "expr": "x^2 + 3*x + 1", "symbol": "x" } }, "confidence": 0.95, "reasoning": "User requests differentiation of polynomial expression" } ``` **Automatic Execution:** The parsed tool call is automatically executed, returning: `2*x + 3` ### 2. Physics Problem **Natural Language Input:** > "Plot the electric field of a point charge at the origin from -5 to 5 meters" **NLI Processing:** ```json { "jsonrpc": "2.0", "id": "2", "method": "nli_parse", "params": { "text": "Plot the electric field of a point charge at the origin from -5 to 5 meters" } } ``` **Expected Tool Call:** ```json { "method": "plot", "params": { "plot_type": "function_2d", "f": "k*q/x^2", "x_range": [-5, 5], "title": "Electric Field vs Distance", "xlabel": "Distance (m)", "ylabel": "Electric Field (N/C)" } } ``` ### 3. Unit Conversion **Natural Language Input:** > "Convert 100 miles per hour to meters per second" **NLI Processing:** ```json { "jsonrpc": "2.0", "id": "3", "method": "nli_parse", "params": { "text": "Convert 100 miles per hour to meters per second" } } ``` **Expected Tool Call:** ```json { "method": "units_convert", "params": { "quantity": {"value": 100, "unit": "mph"}, "to": "m/s" } } ``` ### 4. Complex Multi-Step Problem **Natural Language Input:** > "Solve the differential equation y double prime plus y equals zero with initial conditions y(0)=0 and y'(0)=1, then plot the solution" **NLI Processing:** This complex request gets parsed into multiple tool calls: 1. **Solve ODE:** ```json { "method": "cas", "params": { "action": "solve_ode", "ode": "y'' + y = 0", "symbol": "x", "func": "y", "ics": {"y(0)": 0, "y'(0)": 1} } } ``` 2. **Plot Solution:** ```json { "method": "plot", "params": { "plot_type": "function_2d", "f": "sin(x)", "x_range": [0, 10], "title": "Solution to y'' + y = 0", "xlabel": "x", "ylabel": "y(x)" } } ``` ## NLI Features ### Intent Recognition - **Mathematical Operations**: differentiate, integrate, solve, evaluate - **Plotting Requests**: plot, graph, visualize, show - **Unit Conversions**: convert, transform, change units - **Physical Constants**: get, find, lookup constants - **Data Analysis**: analyze, process, filter signals ### Context Awareness - **Variable Extraction**: Identifies mathematical expressions and variables - **Unit Detection**: Recognizes physical units in natural language - **Parameter Inference**: Fills in reasonable defaults for missing parameters - **Ambiguity Resolution**: Asks for clarification when needed ### Error Handling - **Graceful Degradation**: Falls back to simpler parsing if complex NLI fails - **Suggestion System**: Provides alternative interpretations - **Learning**: Improves over time with usage patterns ## Performance Optimization ### Local LM Benefits - **Low Latency**: ~100-500ms vs 1-3s for cloud APIs - **Privacy**: All processing stays local - **Cost**: No per-token charges - **Availability**: Works offline ### Caching - **Intent Patterns**: Common requests cached for instant response - **Tool Mappings**: Frequent NL→tool mappings stored - **Context Memory**: Recent conversation context maintained ## Advanced Usage ### Chained Operations > "Find the roots of x^3 - 2x^2 + x - 2 = 0, then plot the function and mark the roots" ### Physics Workflows > "Calculate the trajectory of a projectile with initial velocity 50 m/s at 30 degrees, including air resistance" ### Data Analysis > "Load the signal data from experiment.csv, apply a low-pass filter at 100 Hz, then generate a spectrogram" ## Troubleshooting ### Common Issues 1. **LM Server Not Running**: NLI falls back to rule-based parsing 2. **Ambiguous Requests**: System asks for clarification 3. **Unsupported Operations**: Suggests alternative approaches ### Best Practices - Be specific about units and ranges - Use standard mathematical notation - Provide context for complex problems - Check results for reasonableness ## Extensions The NLI system can be extended with: - Domain-specific vocabularies - Custom physics problem templates - Integration with experimental data formats - Multi-language support - Voice input processing