Compute MCP
An MCP server for evaluating arithmetic expressions using a Pratt parser in Rust.
๐ฎ Compute MCP
A robust arithmetic expression evaluator implemented as an MCP (Model Context Protocol) server. This project demonstrates production-quality parser development using modern Rust techniques, comprehensive property-based testing, and adversarial test strategies.
โจ Features
- Complete Arithmetic:
+,-,*,/with correct precedence - Scientific Notation:
1e10,2.5e-3,1.23E+4 - Parentheses Grouping:
(2 + 3) * 4 - Decimal Numbers:
3.14159,-0.5 - Unary Operators:
-42,-(5 + 3),--5 - Robust Error Handling: Division by zero, malformed input, parse errors
- Deep Nesting Support: Handles complex nested expressions
- Property-Based Tested: 60+ tests covering mathematical invariants
๐๏ธ Architecture
Built using a modern Pratt parser for clean operator precedence handling:
Input String โ Pest Grammar โ Pratt Parser โ AST โ Evaluator โ Result
โ โ โ โ โ โ
"2 + 3 * 4" compute.pest PrattParser Expr::Add eval_expr Ok(14.0)
/ \
Expr::Num(2) Expr::Mul
/ \
Expr::Num(3) Expr::Num(4)
Key Components
- Grammar (
src/compute.pest) - Defines syntax with atoms and operators - Pratt Parser (
src/lib.rs) - Handles precedence automatically - AST (
Exprenum) - Immutable expression tree - Evaluator (
eval_expr) - Stack-safe recursive evaluation - MCP Server (
src/bin/stdio_direct.rs) - JSON-RPC interface
๐ Quick Start
Installation
cargo build --release
Direct Evaluation
# Command line tool
cargo run --bin stdio_direct -- eval "2 + 3 * 4"
cargo run --bin stdio_direct -- eval "1e10 / (2.5 + 3.7)"
MCP Server
# Initialize server
echo '{"jsonrpc":"2.0","method":"initialize","params":{},"id":1}' | cargo run --bin stdio_direct
# Batch evaluation
echo '{"jsonrpc":"2.0","method":"tools/call","params":{"name":"evaluate_batch","arguments":{"expressions":["2+2","1e3*2","(5-3)^2"]}},"id":2}' | cargo run --bin stdio_direct
Claude Desktop Integration
{
"mcpServers": {
"compute": {
"command": "/path/to/compute-mcp/target/release/stdio_direct"
}
}
}
๐งช Comprehensive Testing
This project features one of the most thorough test suites for arithmetic parsers:
Test Categories
๐ง Unit Tests (tests/tests.rs)
- Basic arithmetic operations
- Precedence and associativity
- Error handling
- Round-trip parsing
โก Adversarial Tests (tests/adversarial_tests.rs)
- Floating-point edge cases (infinity, NaN, subnormals)
- Deep nesting (1000+ levels)
- Malformed input fuzzing
- Performance stress testing
๐ฏ Property-Based Tests (tests/proptest_adversarial.rs)
- Mathematical invariants (commutativity, associativity, distributivity)
- Parser robustness (never panics)
- Evaluation determinism
- Precision preservation
Key Invariants Tested
// Precedence preservation
parse("a + b * c") == Add(a, Mul(b, c))
// Evaluation determinism
eval(expr) == eval(expr) // Always same result
// Mathematical laws
eval(Add(a, b)) โ eval(Add(b, a)) // Commutativity
eval(Add(Add(a, b), c)) โ eval(Add(a, Add(b, c))) // Associativity
// Round-trip consistency
eval(parse(print(ast))) โ eval(ast)
// Error containment
parse(invalid_input) == Err(_) // Never panics
Running Tests
# All tests
cargo test
# Specific test suites
cargo test --test tests # Basic functionality
cargo test --test adversarial_tests # Edge cases
cargo test --test proptest_adversarial # Property tests
# Parallel execution
cargo test -- --test-threads=4
๐ Bugs Found & Fixed
Property-based testing discovered critical issues during development:
Grammar Ambiguity: The original grammar allowed -5 to parse as either Neg(Number(5)) or Number(-5), causing non-deterministic behavior. Fixed by removing minus signs from number literals.
Precision Edge Cases: Tests revealed floating-point precision issues with expressions like 1e20 + 1 - 1e20, leading to more robust error tolerance.
Deep Nesting Limits: Found parser performance cliffs at ~40+ nesting levels, optimized for practical use cases.
๐ Code Examples
Basic Usage
use compute_mcp::*;
// Simple evaluation
let result = evaluate("2 + 3 * 4")?;
assert_eq!(result, 14.0);
// Scientific notation
let result = evaluate("1.5e3 + 2.5e2")?;
assert_eq!(result, 1750.0);
// Complex expressions
let result = evaluate("((1 + 2) * 3 - 4) / 2")?;
assert_eq!(result, 2.5);
Advanced Features
// Parse to AST for inspection
let ast = parse_expression("-(2 + 3) * 4")?;
// Returns: Mul(Neg(Add(Number(2), Number(3))), Number(4))
// Batch processing
let expressions = vec!["1+1", "2*2", "3/3"];
let results = evaluate_batch(&expressions);
// Error handling
match evaluate("10 / 0") {
Err(ComputeError::DivisionByZero) => println!("Caught division by zero"),
_ => unreachable!(),
}
๐ Project Structure
compute-mcp/
โโโ Cargo.toml # Dependencies and metadata
โโโ src/
โ โโโ lib.rs # Parser, AST, and evaluator (~350 lines)
โ โโโ compute.pest # Pratt parser grammar (~35 lines)
โ โโโ bin/
โ โโโ stdio_direct.rs # MCP server implementation
โโโ tests/
โ โโโ tests.rs # Unit and integration tests
โ โโโ adversarial_tests.rs # Edge case and stress tests
โ โโโ proptest_adversarial.rs # Property-based tests
โ โโโ *.proptest-regressions # Saved failing test cases
โโโ target/ # Build artifacts
๐ Dependencies
[dependencies]
pest = "2.6" # Parser generator
pest_derive = "2.6" # Derive macros for grammar
lazy_static = "1.4" # Global parser instance
mcpr = "0.2.3" # MCP protocol
serde = "1.0" # JSON serialization
clap = "4.4" # Command line interface
[dev-dependencies]
proptest = "1.6.0" # Property-based testing
๐ Educational Value
This project serves as an excellent case study for:
- Modern Parser Design: Pratt parsers vs recursive descent
- Property-Based Testing: Discovering edge cases automatically
- Rust Best Practices: Error handling, type safety, zero-cost abstractions
- Protocol Implementation: MCP server development
- Mathematical Correctness: Ensuring arithmetic laws hold
Perfect for blog posts, tutorials, and educational content about robust software development.
๐ License
MIT - See LICENSE for details.
๐ฎ Ready for production use with confidence backed by comprehensive testing! ๐ฎ
Related Servers
Serena
A coding agent toolkit that provides an LLM with IDE-like capabilities for semantic code retrieval and editing directly in your codebase.
Assistant MCP Server
An MCP server that dynamically loads tools from an external JSON file configured via an environment variable.
MCP Inspector
A developer tool for testing and debugging MCP servers with a web-based UI.
Kibana MCP Server
Access and interact with your Kibana instance using natural language or programmatic requests.
Mentor MCP
Provides AI-powered mentorship to LLM agents for tasks like code review, design critique, and brainstorming, using the Deepseek API.
Build-Scout
Interact with various build systems including Gradle, Maven, NPM/Yarn, Cargo, Python, Makefile, and CMake.
Apifox MCP Server
Provides API documentation from Apifox projects as a data source for AI programming tools that support MCP.
nf-core MCP Server
Manage and navigate local nf-core pipeline repositories.
FAL Imagen 4
Generate high-quality images using Google's Imagen 4 Ultra model via the FAL AI platform.
Note MCP
MCP servers for Notecard, Notehub, and general development.