Sedna Labs is an independent AI lab building practical Rust infrastructure for autonomous agents and the humans who operate them.
Our public work focuses on Model Context Protocol tooling, terminal-native agent workflows, operator-controlled automation, local credential control, and small observability systems that make complex environments easier to inspect and verify.
We start from a simple assumption: AI agents are useful because they move quickly, and risky because they make mistakes. They should not be trusted to produce secure systems or operate safely by default.
The goal is not to slow agents down. The goal is to make fast agent work survivable: bounded actions, clear provenance, human review at the right points, audit trails, and fail-closed paths when the system cannot prove what happened.
- Rust MCP foundations: mcp-toolkit-rs provides reusable building blocks for MCP servers and clients, with an emphasis on clear contracts, local control, testability, and observable behavior.
- Operator-controlled MCPs: cloudflare-mcp, postgres-mcp, google-admin-mcp, and ga4-mcp make operational surfaces inspectable and auditable through agent-facing tools.
- Computer-use and agent runtime: android-computer-use-mcp exposes local Android emulator observation and control, while codex carries terminal-first coding-agent workflows.
- Focused systems tools: moshwatch reflects the same bias toward precise instruments, observable systems, and practical operator workflows.
We prefer small, composable tools with clear boundaries, local control, contract tests, and verification paths that humans can audit.