Pinned Tweet
$CAPA is LIVE on @RobinhoodApp.
CA: 0xb54ef4bd0a97083862e81a97d13af14594f4203d
The open-source capabilities manager built to unify AI coding agents.
- Manage skills and tools.
- Connect MCP servers.
- Simplify agent configs.
One setup. Every agent.
$CAPA Dev remains active!
Another buyback and supply burn completed. We continue supporting $CAPA, reducing supply, and staying committed to keeping the chart strong for the long term.
Consistent actions, long-term vision.
robinhoodchain.blockscout.co…
Prefer moving cards to editing YAML?
capa's local web UI lays out skills, tools, rules and hooks you can drag into order, guides OAuth and secret setup, and lets you browse registries. Each change writes straight back to capabilities.yaml.
3
3
261
Clip a probe onto any MCP server.
capa sh calls a tool straight from your terminal and shows the result, using your agents' credentials. Turn the dial with --raw to read the payload before any formatter.
$CAPA Buyback & Burn continues!
We’ve bought back more $CAPA and burned additional supply, staying active and committed to supporting the chart for the long term.
Building stronger, moving forward.
robinhoodchain.blockscout.co…
Play in the sandbox, keep the house clean.
capa wrap builds .cursor/, .claude/, skills and MCP configs inside ~/.capa/workspaces while your repo stays untouched. Finished experimenting? capa wrap --prune rakes it all away.
Onboarding a new dev shouldn't include an afternoon of copying agent configs.
capa's lockfile pins every skill and MCP server to an exact SHA, so the person who clones your repo tomorrow gets the exact agent setup you have today.
Think of your agent setup as a relay baton.
capabilities.lock carries every pinned skill and MCP server from you to the next teammate to your CI machine. Nobody drops it, nobody rebuilds it. Clone, capa install, keep running.
New coding agent every month? Keep your setup.
Add the agent to providers in capabilities.yaml, run capa install, and it gets the same skills, MCP servers and rules in its own native format. 35+ agents supported.
When an MCP tool misbehaves, you shouldn't need a whole agent session to find out why.
capa sh calls any configured tool straight from the terminal, using the same credentials. Add --raw to see exactly what came back.
Want to try a new skill or MCP server without leaving a mess in your repo?
capa wrap runs your agent from a shadow workspace in ~/.capa/workspaces. Experiment freely, then capa wrap --prune clears out stale workspaces.
Not every registry is just a list.
Most capa registry adapters are executable TypeScript, so read the source before you enable a third-party one. Claude marketplace catalogs are plain JSON, which makes them safer by design.
$CAPA Dev is active again! 🔥
We’re back with another buyback and supply burn to keep $CAPA moving forward, strengthen the project, and support long-term growth.
Building with consistency. Staying committed for the long run.
robinhoodchain.blockscout.co…
3
8
265
Front-loading every MCP tool fills your agent's context before it starts working.
Set toolExposure to on-demand and capa hands tools over only when a task needs them. capa's own benchmark: 19–40% fewer tokens over 150 trials.
6
5
290
Would you give your research agent the keys to production?
With capa, every sub-agent gets its own filtered MCP endpoint. The research agent searches and reads. The release agent pushes and deploys. Nobody gets a door they don't need.
1
2
160
The biggest toolset in capa's benchmark: 108 GitHub tools.
All three ways of exposing them scored 100%. Going through capa sh was cheapest at $0.1465 a run, vs $0.1596 with everything exposed.
capa.sh/benchmarks







