Arcade’s engineering team is constantly building and constantly pushing on what agents can do. Arcade Labs is where the latest of that work lives in public: research, prototypes, building blocks, and the opinionated little tools we made for ourselves and figured other people will want too. New work lands here the moment it’s worth sharing.
A few ground rules: nothing in Arcade Labs is supported, none of it is a roadmap commitment, and breaking changes can happen on any commit. Some projects will graduate into the Arcade product. Some will be archived.
Find Arcade Labs at github.com/arcadeai-labs.
Launching today: Omni
Omni is an MCP server pre-loaded with 500+ optimized tools. Google Workspace, Microsoft 365, Zendesk, Twitter, and the rest of the daily-driver SaaS. All pre-tested, all ready, all on a single connection.
Most agent setups today force you to load every tool schema into the model up front, burning tens of thousands of tokens on definitions the model will never use in a given turn. Omni flips that: your agent sees a tiny surface with three tools:
- Search
- Call
- Authorize
and pulls in exactly the tools it needs the moment it needs them, ranked by semantic relevance to what the user actually asked for. One line of config in Claude Code, Cursor, Claude Desktop, or any MCP client, and your agent has every tool in the catalog at its fingertips with none of the bloat.
What makes Omni compelling isn’t just efficiency — it’s that the plumbing disappears. The agent doesn’t narrate “let me search for the right tool,” doesn’t show the user a picker, and doesn’t ask which integration to use. The user asks for an outcome and the agent delivers it, reaching for Slack, GitHub, Gmail, Calendar, or anything else in the catalog as naturally as a person would. When a tool needs authorization, Omni handles the OAuth handshake, the account switching, and the scope repair without dropping back to the user for tribal knowledge. The result feels less like an agent wired to a fixed set of integrations and more like an agent that genuinely knows how to use the internet.
Omni is live at `https://omni.arcade.dev`. Point your favorite MCP client at it, sign in once, and the catalog is yours. If you’re experimenting with MCP and want to see what dynamic tool discovery actually feels like in practice, Omni is the easiest way to find out.
Also in Labs: Agent Library
Agent Library is our local-first, MCP-native memory layer for agents, and it lives in Arcade Labs too. Agent Library runs as a SQLite file you can hold in your hand: hybrid semantic + keyword search, read-mostly by design, runs on your own machine.
Watch the walkthrough video, grab the code at github.com/arcadeai-labs/agent-library, or read the original announcement for the full design rationale.
What’s next
More. We’re constantly building, and our default from here is to share. Watch the org, try the things, open issues when something breaks. Welcome to Arcade Labs.
FAQ
What is Arcade Labs?
Arcade Labs is Arcade’s public space for research projects, prototypes, and experiments from the Arcade engineering team. Arcade Labs lives at github.com/arcadeai-labs. Projects in Arcade Labs are not part of the supported Arcade product.
How is Arcade Labs different from the Arcade product?
The Arcade product is supported, versioned, and covered by SLAs. Arcade Labs is not. Projects in Arcade Labs are provided AS-IS, with no roadmap commitments, no support agreements, and no guarantees that breaking changes won’t land on any commit. Some Arcade Labs projects will eventually graduate into the Arcade product; most will not.
What is Omni?
Omni is an MCP server pre-loaded with over 500 productivity and work tools, including Google Workspace, Microsoft 365, Zendesk, and Twitter. Omni connects to any MCP client (such as Claude) at https://omni.arcade.dev/mcp and exposes its tools through on-demand discovery rather than loading every definition into the agent’s context window upfront.
Which tools are included in Omni?
Omni launches with a curated set of over 500 pre-tested productivity, media, and work tools across the most commonly used SaaS: Google Workspace, Microsoft 365, Zendesk, Twitter, and similar daily-driver categories. The set is fixed at launch; Omni does not currently support adding custom tools.
What is Agent Library?
Agent Library is Arcade’s local-first, MCP-native memory layer for AI agents. Agent Library runs as a local SQLite file with hybrid semantic + keyword search, ingests text, code, PDFs, and images, and is read-mostly by design so an agent can search the library without corrupting it. The source code is at github.com/arcadeai-labs/agent-library.
Are Arcade Labs projects open source?
Some are. Each project in Arcade Labs sets its own license. Open-source repositories in Arcade Labs typically use the Apache 2.0 license, but specific terms vary by project. Some Arcade Labs projects, such as Omni, are hosted services rather than open-source code repositories.
Where do I file bugs or ask questions about Arcade Labs?
File GitHub issues on the relevant Arcade Labs repository. The Arcade engineering team reviews issues as time permits, but there are no formal SLAs for Arcade Labs projects. Do not submit Arcade support tickets for Arcade Labs issues.
