kodegen-mcp-schema

Lightweight MCP (Model Context Protocol) tool schema definitions for the kodegen ecosystem.

Overview

kodegen_mcp_schema provides type-safe schema definitions for AI agent tools and interactions. This library serves as the single source of truth for all tool schemas, with minimal dependencies and maximum reusability.

Design Philosophy: Schema-only library with zero heavy dependencies. Contains only Args and PromptArgs type definitions using serde, schemars, and serde_json.

Installation

Add this to your Cargo.toml:

[dependencies]
kodegen_mcp_schema = "0.1.0"

Or install directly from the repository:

[dependencies]
kodegen_mcp_schema = { git = "https://github.com/cyrup-ai/kodegen-mcp-schema" }

Features

Comprehensive Tool Schemas

Filesystem - File operations, search, directory management
Terminal - Command execution and process control
Git - Version control operations (init, commit, branch, worktree)
GitHub - Issues, PRs, code search, security scanning
Browser - Web automation, research sessions, agent-based browsing
Database - Schema inspection, SQL execution, connection pooling
Claude Agent - Multi-agent spawning with memory and reasoning
Web Scraping - Crawling, content extraction, search indexing
Reasoning - Sequential thinking, MCTS, beam search strategies
Prompts - Template management and rendering
Introspection - Usage stats and tool call history

Type-Safe Schemas

All schemas use Rust's type system with:

JSON Schema generation via schemars
Serde serialization/deserialization
Validation attributes for constraints
Comprehensive documentation

Usage

use kodegen_mcp_schema::{ReadFileArgs, StartSearchArgs, SearchType};

// File reading with optional pagination
let read_args = ReadFileArgs {
    path: "/path/to/file.txt".to_string(),
    offset: 0,
    length: Some(100),
    is_url: false,
};

// Advanced file search
let search_args = StartSearchArgs {
    path: "/project".to_string(),
    pattern: "TODO".to_string(),
    search_type: SearchType::Content,
    file_pattern: Some("*.rs".to_string()),
    max_results: Some(100),
    ..Default::default()
};

// GitHub operations
use kodegen_mcp_schema::{CreateIssueArgs, SearchCodeArgs};

let issue = CreateIssueArgs {
    owner: "cyrup-ai".to_string(),
    repo: "kodegen-mcp-schema".to_string(),
    title: "Bug report".to_string(),
    body: Some("Description here".to_string()),
    labels: Some(vec!["bug".to_string()]),
    assignees: None,
};

Schema Pattern

Each tool follows a consistent pattern with two primary types:

XxxArgs - Runtime arguments for tool execution
XxxPromptArgs - Arguments for generating contextual prompts

Example:

#[derive(Debug, Clone, Serialize, Deserialize, JsonSchema)]
pub struct ReadFileArgs {
    pub path: String,
    #[serde(default)]
    pub offset: i64,
    #[serde(default)]
    pub length: Option<usize>,
}

Development

Prerequisites

Rust nightly toolchain
Components: rustfmt, clippy

Build and Test

# Build the library
cargo build

# Run tests
cargo test

# Run clippy linter
cargo clippy

# Format code
cargo fmt

# Check without building
cargo check

Architecture

Module Organization

The library is organized into domain-specific modules:

src/
├── lib.rs              # Re-exports all types
├── filesystem.rs       # File operations and search
├── terminal.rs         # Process and command execution
├── git.rs             # Git operations
├── github.rs          # GitHub API interactions
├── browser.rs         # Browser automation
├── citescrape.rs      # Web crawling
├── database.rs        # Database operations
├── claude_agent.rs    # Agent spawning and memory
├── reasoning.rs       # AI reasoning tools
├── prompt.rs          # Template management
├── config.rs          # Configuration
├── process.rs         # System processes
└── introspection.rs   # Usage statistics

Session-Based Patterns

Several tools use async session patterns for long-running operations:

Browser Research: start_browser_research → get_research_status → get_research_result
File Search: start_search → get_search_results → stop_search
Web Crawling: scrape_url → scrape_check_results → scrape_search_results
Terminal Commands: start_terminal_command → read_terminal_output → send_terminal_input
Claude Agents: spawn_claude_agent → read_claude_agent_output → send_claude_agent_prompt

Advanced Features

Powerful Search Capabilities

The filesystem search tool supports:

Multiple search modes (files, content)
Regex engines (Rust, PCRE2)
Case modes (sensitive, insensitive, smart)
Boundary matching (word, line)
Output formats (full, files-only, count-per-file)
Sorting, preprocessing, and compression handling

Claude Agent Memory

Memory tools for persistent agent knowledge:

Memorize - Store information in named libraries
Recall - Semantic search across memories
List Libraries - View available memory stores
Check Status - Monitor async memorization

Contributing

Contributions are welcome! Please ensure:

Code follows existing patterns (Args/PromptArgs structure)
All types derive required traits (Debug, Clone, Serialize, Deserialize, JsonSchema)
Use #[serde(default)] for optional fields with defaults
Add comprehensive documentation
Run cargo fmt and cargo clippy before submitting

License

This project is licensed under the MIT License - see the LICENSE file for details.

Keywords

ai, ml, agents, candle, inference, mcp, schemas, tools, automation

kodegen_mcp_schema 0.10.14