Agent infrastructure, MCP servers, and developer tools for teams that need AI workflows to run in real environments.
We build small systems around strict tool contracts, searchable operational memory, browser-visible validation, and review loops that can survive production constraints.
- Agent Capability Router - a runtime-neutral skill that helps Codex and Claude Code choose when to use tools, plugins, skills, subagents, automations, and deeper research.
- Visual Eyes - visual inspection for running web apps in Claude Code.
- Semtree - semantic code-tree tooling for assistants navigating large repositories.
- TranscreveAI - video links, Reels, frames, OCR, and transcripts into multimodal dossiers, content packs, skill drafts, batch reports, and RAG-ready knowledge.
- MCP Fiscal Brasil - Brazilian fiscal workflows through MCP: CNPJ, NFe, NFSe, SPED, eSocial, and related automations.
- ShieldCode - security hardening and production-grade error handling for AI-assisted development.
- claude-orchestration - multi-agent orchestration guidelines and reusable operating patterns.
- agent-stack-public - public map of a Codex and Claude Code agent stack, sanitized for reuse.
- Tool surfaces agents can call safely.
- Context that becomes reusable memory.
- CI, review gates, and traceable decisions.
- Browser/UI inspection as part of coding-agent workflows.
- Brazilian business, fiscal, and operational automation.
TranscreveAI is the current center of gravity: an agent-ready video intelligence pipeline that can process messy social/video inputs, extract evidence, generate reusable artifacts, index the result, and hand it back to a parent agent as durable knowledge.