The Exchange
The Invoked Exchange is a free, hand-curated directory of MCP surfaces, agent skills, and tools with real performance data. Search it in-app or on the web, build a loadout, and invoke resources into your workspace in one click.
The Invoked Exchange is a free, hand-curated public directory of MCP surfaces, agent skills, and tools — each with real-world performance data. It is where you discover what to connect Invoked to, and it works both on the web and inside the desktop app.
Everything in the Exchange is 100% free to browse and curated by developer advocates.
What's in the Exchange
| Resource | What it is |
|---|---|
| Surfaces | MCP servers and APIs you can add to a workspace |
| Skills | Reusable agent behaviors (recipes) you can drop in |
| Tools | Individual capabilities exposed by a surface |
Public pages live at invoked.ai/exchange, with per-resource pages like /surfaces/<slug> and /skills/<slug>. Each page is agent-readable — the same content is available as Markdown for LLM crawlers.
Searching in the app
Open Explore → Exchange in the sidebar (or press ⌘K and search "Exchange"). Type to filter the catalog by name or description, toggle between Surfaces and Skills, and add what you need:
- Add a surface — pick it, supply any required configuration, and it's added to your workspace.
- Add a skill — one click drops it into your skill library.
Search is instant: the catalog loads once and filtering happens locally.
Loadouts
A loadout is a staging set you gather across searches and add all at once. Use the + on any result to stage it, then Add to workspace to apply the whole set — skills and configuration-free surfaces are added immediately; surfaces that need credentials are kept back so you can finish configuring them.
Contributing performance data
The Exchange gets better as people use it. If you opt in, Invoked contributes anonymous, performance-only metrics (success rate and latency) for the catalog surfaces you run, so others can see what actually works.
What's shared: whether calls succeed, how long they take, a coarse error category, and the model family. Never shared: your prompts, data, API keys, exact tools, or identity. Contribution is off by default, opt-in, and limited to signed-in workspaces using catalog surfaces. Toggle it any time in Settings → Privacy.
Aggregated metrics only appear publicly once enough independent workspaces have contributed, so no single workspace is ever identifiable.