Plug your AI agent into the REAA Standard Library

Today we're shipping the REAA Standard Library (SL) as a distribution surface for AI coding agents. Every agent that supports the Model Context Protocol — Claude Code, Cursor, Codex, Windsurf, Cline, Gemini — can now reach hundreds of MIT-licensed TypeScript packages with one terminal command. Chat surfaces without MCP support get the same catalog as static files at the root of the domain.

One line for Claude Code

From any terminal:

claude mcp add --transport http reaatech-sl https://mcp.reaatech.com/sl

Open a Claude Code session, type /mcp, and you'll see reaatech-sl with a green dot. The agent can now search the catalog, pull package details, list categories, and file issues on the package repos when something breaks — all from inside the session, without leaving the chat.

For Cursor, Codex CLI, Windsurf, Cline, Gemini CLI, or any chat surface that doesn't run MCP, the setup page has per-agent snippets.

What's in the catalog

Hundreds of MIT-licensed TypeScript packages across seven categories, all built and maintained in the open at github.com/reaatech:

• Testing & Security — chaos testing, prompt-injection benches, tool-use firewalls.
• Evals & Quality — agent and RAG eval harnesses, LLM-as-judge, prompt versioning.
• Observability & Cost — OTel GenAI semconv, cost telemetry, semantic caching, model routing.
• Orchestration & Protocols — agent mesh, confidence routing, handoff protocols, A2A.
• MCP Infrastructure — gateways, catalogs, doctor tooling, schema evolution, server starters, multi-tenancy.
• Reliability & Ops — circuit breakers, idempotency, output repair, secret rotation, session continuity.
• Domain Pipelines — hybrid RAG, voice agent kit, media pipelines, agent memory.

You can browse and filter the whole catalog with search, category filters, repo filters, and a grid-or-list view toggle. The catalog is regenerated continuously as we ship new packages.

Two paths in

The MCP server at mcp.reaatech.com/sl is the preferred path for any agent that supports MCP. It exposes five tools:

• list_categories — orientation. The seven categories with package counts.
• list_by_category — every package in a category, grouped by repo.
• get_package — full details for a single package by name.
• search_packages — natural-language search across the catalog.
• report_issue — files a deduplicated GitHub issue on the package's repo when something breaks (see below).

For agents without MCP support — older clients, chat-only surfaces, custom setups — the same catalog lives in two static files at the root of the domain:

• llms.txt — short discovery file (~3KB), per the llms.txt convention.
• llms-full.txt — comprehensive per-package dump (~200KB), suitable for pasting into any chat with web fetch.

Agents file their own bugs

The most useful tool on the MCP server is probably the one nobody asks for upfront: report_issue.

When an agent hits a problem while using a REAA package — wrong type signature, broken example in the README, a runtime crash from an edge case — it can call report_issue with the package name and the error context. The server scrubs PII (emails, home directory paths, suspected secrets) into redacted markers, computes a dedup hash, and either files a new GitHub issue on the package's repo or appends a +1 comment to an existing one with the same hash.

The result: real usage friction surfaces as deduplicated, labeled issues in the right repo, automatically — no manual triage of "where does this go," no inbox to monitor.

What's open vs. hosted

The packages are MIT-licensed; you can drop them into your stack and never talk to us. The MCP server, the catalog browser, the auto-generated skills repo for agents using progressive-disclosure skill files, the per-agent setup snippets — all of those are free too. Use what you need; we don't track you.

What we offer on top of the open source is consulting, implementation, and hosted solutions for teams who'd rather have us assemble the right packages into a working system for their business. Book a conversation if that's you.

Coming next

The MCP server is operational and stable. The polish items still in flight:

• Semantic search via a bundled embedding index. The current search_packages is lexical — it does well on direct queries and less well when the natural-language description doesn't share vocabulary with the package's name or tags.
• Rate limiting on report_issue per agent identity + IP, once we have real traffic data to size it.
• OpenTelemetry tool-call spans + alerting on the live server.
• Per-repo editorial copy across all of our SKILL.md files, currently rolling out through an admin review queue.

None of these are user-facing blockers. They sharpen the edges.

Start here

• Setup page — per-agent install snippets.
• Browse the catalog — search, filter, drill in.
• github.com/reaatech — every package's source.
• Book a conversation if you'd rather have us wire it in.