A high-performance string search utility with MCP (Model Context Protocol) integration for the infinity-topos environment. This is a wrapper around krep, an ultra-fast pattern matching utility that significantly outperforms traditional tools like grep.
THE KREP-MCP-SERVER ABSURDITY DIAGRAM
====================================
+-----------------------------------------------------+
| |
| The KREP MCP Redundancy Zone |
| |
+-----^--------------------+------------------------+-+
| | |
+-----------+----------+ +------+-------------+ +------+-------+
| | | | | |
| M E T A D A T A | | F U N C T I O N | | B I N A R Y |
| E X P L O S I O N | | N A M E C H A O S| | H U N T |
| | | | | |
+-----+----------+-----+ +---+-----------+----+ +------+-------+
| | | | |
v v v v v
+--------+--+ +----+-----+ +---+----+ +---+-----+ +----+------+
| | | | | | | | | |
| "Unified" | | 37 Paths | | krep | |krepSearch| | 5 Error |
| Function | | To Find | | | |krepMatch | | Handlers |
| That Does | | The Same | | | |krepCount | | For |
| 3 Things | | Binary | | | | | | 1 Error |
| | | | | | | | | |
+-----------+ +----------+ +--------+ +----------+ +-----------+
+-----------------------------------------------------+
| |
| Configuration & Shell Script Hell |
| |
+-----^--------------------+------------------------+-+
| | |
+-----------+----------+ +------+-------------+ +------+-------+
| | | | | |
| 3 Scripts to | | Integer | | Test Mode |
| Install 1 Thing | | Arithmetic in | | that Mocks |
| | | Shell that says | | Success When|
| | | 0 + 0 = Syntax | | Everything |
| | | Error | | Fails |
+----------------------+ +--------------------+ +--------------+
"It's not redundant if it's resilient!"
- MCP Engineer, probably
Krep MCP Server provides a unified interface to the krep binary, a high-performance string search utility similar to grep but with optimized algorithms and multi-threading capabilities. It exposes krep's functionality through the Model Context Protocol, allowing AI assistants to perform efficient pattern searching in files and strings.
- High-Performance Search: Uses optimized algorithms (KMP, Boyer-Moore-Horspool, Rabin-Karp) selected based on pattern length
- Hardware Acceleration: Leverages SIMD instructions (SSE4.2/AVX2 on x86/x64, NEON on ARM) when available
- Optimized Multi-Threading: Automatically uses all available CPU cores for maximum parallel search performance
- Unified Interface: Single function with multiple modes (file search, string search, count-only)
- MCP Integration: Seamless integration with AI assistants through the Model Context Protocol
This codebase demonstrates how a simple tool (a wrapper for a string search utility) became bloated with unnecessary complexity:
-
Simple Core, Complex Implementation: The actual functionality is straightforward but buried under layers of over-engineering
-
Documentation Overload: 15 documentation files for a tool that could be explained in a single well-structured README
-
Integration Madness: 3 separate integration systems (Cline, Claude Desktop, SDK), each with redundant scripts and documentation
-
Installation Script Proliferation: 7 installation scripts when one configurable script would suffice
-
Error Handling Duplication: Error handling duplicated at multiple levels rather than having a unified approach
-
Test Fragmentation: Test files scattered across the codebase rather than being organized systematically
-
Configuration Redundancy: Configuration files and environment variables duplicated across multiple components
-
Binary Path Overkill: Searches 37 different paths for a single binary that should be in one predictable location
What It Should Have Been:
┌──────────────────────┐
│ krep-mcp-server │
│ ┌────────────────┐ │
│ │ index.js │ │
│ │ - one function│ │
│ └────────────────┘ │
│ ┌────────────────┐ │
│ │ README.md │ │
│ │ - clear docs │ │
│ └────────────────┘ │
│ ┌────────────────┐ │
│ │ install.sh │ │
│ │ - one script │ │
│ └────────────────┘ │
└──────────────────────┘
Here's the actual project structure:
krep-mcp-server/
├── CLINE_README.md
├── CLINE_SETUP.md
├── CLAUDE_DESKTOP_INTEGRATION.md
├── CLAUDE_DESKTOP_README.md
├── EXAMPLES.md
├── IMPLEMENTATION_SUMMARY.md
├── INSTALL_NOW.md
├── LIFECYCLE_DESIGN.md
├── MCP_COMPLIANCE.md
├── MCP_URIS.md
├── README.md
├── SETUP_CLAUDE_DESKTOP.md
├── TESTING_STRATEGY.md
├── THREAD_OPTIMIZATION.md
├── analysis/
│ └── index.tree.json
├── auto-install-claude.sh
├── cline-config.js
├── direct-install.sh
├── eslint.config.js
├── fix-claude-desktop.sh
├── go-integration/
│ ├── example/
│ └── krep.go
├── install-claude-desktop.sh
├── install-cline-integration.sh
├── install-sdk-integrations.sh
├── jest.config.js
├── just-krep.sh
├── mcp-config.json
├── package-lock.json
├── package.json
├── python-integration/
│ └── krep_mcp_client.py
├── run-claude-desktop.sh
├── run-claude-integration.sh
├── run-cline-mcp-server.sh
├── run-cline-test.sh
├── run-tests.sh
├── run.sh
├── sdk-integration.js
├── src/
│ ├── index.js
│ ├── index.min.js
│ ├── mcp_server.js
│ └── mcp_server.min.js
├── Support/
│ └── Claude/
├── test/
│ ├── benchmark.js
│ ├── fixtures/
│ ├── integration/
│ ├── mcp_benchmark.js
│ ├── mock-server.js
│ ├── unit/
│ └── utils.js
└── various test scripts...
-
Ensure you have the krep binary installed:
cd /path/to/krep-native make
-
Configure the MCP server in your MCP settings file:
{ "mcpServers": { "krep": { "command": "node", "args": [ "/path/to/krep-mcp-server/src/index.js" ], "env": { "CLAUDE_MCP": "true", "KREP_PATH": "/path/to/krep-native/krep", "DEBUG": "true" }, "description": "High-performance string search utility with unified interface", "disabled": false, "autoApprove": [ "krep" ] } } }
The krep MCP server exposes a single unified function:
<use_mcp_tool>
<server_name>krep</server_name>
<tool_name>krep</tool_name>
<arguments>
{
"pattern": "search pattern",
"target": "file path or string to search",
"mode": "file|string|count",
"caseSensitive": true|false,
"threads": null // Automatically uses all CPU cores if not specified
}
</arguments>
</use_mcp_tool>
- pattern (required): The pattern to search for
- target (required): File path or string to search in
- mode (optional): Search mode
file
(default): Search in a filestring
: Search in a stringcount
: Count occurrences only
- caseSensitive (optional): Whether the search is case-sensitive (default: true)
- threads (optional): Number of threads to use (default: auto-detected based on CPU cores)
See examples.md for detailed usage examples and patterns.
The krep MCP server works by:
- Receiving requests through the Model Context Protocol
- Parsing the request parameters
- Building the appropriate krep command based on the mode and parameters
- Executing the command using the krep binary
- Parsing the results and returning them in a structured format
Krep is designed for high-performance pattern searching:
- Algorithm Selection: Automatically selects the optimal algorithm based on pattern length
- KMP (Knuth-Morris-Pratt) for very short patterns (< 3 characters)
- Boyer-Moore-Horspool for medium-length patterns (3-16 characters)
- Rabin-Karp for longer patterns (> 16 characters)
- Hardware Acceleration: Uses SIMD instructions when available
- Dynamic Multi-Threading: Automatically utilizes all available CPU cores for optimal parallel search performance
The krep-mcp-server can be integrated with the Cline VSCode extension, allowing you to use high-performance string search capabilities directly in your VSCode environment.
We provide an automatic installation script to set up the Cline integration:
# Install the integration
./install-cline-integration.sh
# Test the integration before installing
./run-cline-test.sh
# Uninstall the integration
./uninstall-cline-integration.sh
Once integrated, you can use krep directly in Cline conversations:
/krep krep pattern="function" target="/path/to/search" mode="file"
For detailed instructions and usage examples, see:
- CLINE_SETUP.md - Setup instructions
- CLINE_README.md - Usage guide
Krep MCP Server is designed to work seamlessly within the infinity-topos environment:
- Babashka Integration: Use Babashka to process search results
- Say Integration: Vocalize search results using the Say MCP server
- Coin-Flip Integration: Use randomization to determine search strategies
CLAUDE_MCP
: Set to "true" to run in MCP modeKREP_PATH
: Path to the krep binaryDEBUG
: Set to "true" for verbose loggingKREP_TEST_MODE
: Set to "true" to run in test mode with mock responsesKREP_SKIP_CHECK
: Set to "true" to skip checking if the krep binary exists
When not running in MCP mode, the server starts an HTTP server with the following endpoints:
GET /health
: Health check endpointGET /
: Server informationPOST /search
: Search for patterns in filesPOST /match
: Match patterns in stringsGET /performance
: Performance informationGET /algorithm-selection
: Algorithm selection guide
MIT