Kilo Kit MCP

What VoDaiLocz/kilo-kit-mcp MCP server does

VoDaiLocz/kilo-kit-mcp MCP server is an agent runtime and skill catalog for the 🤖 Coding Agents category. It is designed to keep coding work inside a staged workflow: inspect the task and available context, reason about implementation options, obtain approval, execute changes, and verify the result.

The runtime separates planning from low-level actions. Its cognitive layer handles Tree-of-Thoughts branching, adversarial review, 5-Whys root-cause analysis, context handling, and grounded synthesis. Execution is handled through guarded operations such as atomic file writes, AST-based edits, and security-filtered command execution. The README describes 180 curated skills and 24 MCP tools.

How it works

Tasks move through five C4 gates. Gate 1 performs grounded probing through memory recall and codebase inspection. Gate 2 evaluates alternatives with a three-option Tree-of-Thoughts graph, adversarial review, and confidence tracking. Gate 3 locks the plan and injects relevant skills. Gate 4 permits execution under the supervisor, while Gate 5 performs four-dimensional quality checks, including acceptance criteria, clean-code interfaces, UI/UX review, and Playwright browser or DOM verification.

Kilo-Sentinel operates as JSON-RPC interceptor middleware between the host client and the runtime. Its pre-flight lock rejects edits to files that have not been initialized or read, helping enforce planning before mutation. A loop tripwire can react to repeated identical calls or edit thrashing, and a circuit breaker can return the workflow to an earlier stage after repeated loop detection.

Reasoning can be stored in SQLite tables including cognitive_triangulations, katl_trajectories, and facts. The triangulation workflow combines stored memory, GitHub grounding, and a Tree-of-Thoughts benchmark. The README states that confidence below 0.70 triggers research escalation through a subagent.

Tools and capabilities

The documented tool groups include:

• kilo_orchestrate_task for the C4 closed-loop workflow.
• kilo_route_intent for selecting task modes and workflow chains.
• kilo_get_skill and kilo_search_skills for loading and finding skills.
• kilo_memory_report, kilo_remember_fact, and kilo_record_reflection for SQLite-backed memory and lessons.
• kilo_route_report and kilo_validate_skills for routing telemetry and catalog checks.
• kilo_triangulate_research for grounded synthesis and confidence-based escalation.
• kilo_think_step for branching hypotheses and comparing three options.

The complete release is described as a 24-tool MCP suite, although the supplied excerpt lists only part of that suite.

Setup and configuration

The package is published as @vodailocz/kilo-kit-mcp. The README identifies Node.js 20 or newer and TypeScript 5.9 as the runtime and implementation baseline. The architecture diagram shows host clients connecting through MCP over stdio. Configure the package as an MCP stdio server in a compatible client; the supplied material does not provide a complete client-specific configuration block or required environment variables.

Limitations and notes

VoDaiLocz/kilo-kit-mcp MCP server is focused on agent workflow control and coding operations rather than a third-party API integration. The excerpt does not document external service credentials, environment variables, detailed tool schemas, or the exact installation command. Playwright verification is described as part of the quality gate, but the supplied material does not specify browser setup requirements. Review the repository before adopting it where execution policy, persistence, or approval behavior must be customized.