Jev Ultrafast MCP: an autonomous browser for Claude Code
Jev Ultrafast MCP replaces 20 browser calls with a single browser_goal. 10x fewer tokens, 10x cheaper than Claude in Chrome. Full setup with benchmarks.
Command Palette
Search for a command to run...
Articles on development, AI and lessons learned in production.
Jev Ultrafast MCP replaces 20 browser calls with a single browser_goal. 10x fewer tokens, 10x cheaper than Claude in Chrome. Full setup with benchmarks.
A framework at 3M req/s never holds that number in production. The 5 walls that melt throughput, measured on HTTP Arena: JSON, TLS, database, cloud quotas.
Past ~30 skills, an AI agent picks the right tool less and less. The real culprit is not a bloated context but broken routing. Numbers, mechanism, and concrete countermeasures.
Three approaches to connect tools to Claude: JSON schema with manual loop, Tool Runner SDK, and programmatic tool calling. Comparison, code examples, and practical guidance.
CodeRift analyzes every GitLab merge request in 7 steps: diff parsing, tree-sitter AST, TOML rules, AI agents and automatic inline comments.
Entities, store, engine, routes: how to structure a Rust API in strict layers with Axum, SQLx, and utoipa. Code examples from IronFlow in production.
IronFlow is a workflow engine where workflows are imperative Rust code, not YAML. Typed FSM, 10 AI providers, native parallelism with Tokio.
How to structure a SKILL.md, leverage progressive disclosure to save context, and avoid common anti-patterns. A guide with production skill examples.
My real Claude Code setup with the actual files: CLAUDE.md under 30 lines, 15 path-scoped rules, 40 skills and one RTK hook that saved 276M tokens.
MCP RTK is a Rust proxy that filters MCP server responses through an 8-step pipeline, reducing Claude Code's token usage by 60 to 90%.