Give your coding agent the whole stack.
Package MCP servers, APIs, and commands as focused Caplets. Your agent opens only what each task needs.
You choose every capability and credential boundary.
Any backend. One capability layer.
Turn the systems you already use into focused capabilities that travel across supported agents.
CONNECT
- MCP
- OpenAPI
- Google APIs
- GraphQL
- HTTP
- CLI
- Caplet files
CapletREUSE
- CodexMCP
- ClaudeMCP
- OpenCodeNative · MCP
- PiNative · MCP
- Remote clientsRemote Attach
The definition travels. Credentials stay with each host.
Create a CapletOne agent. The whole job.
Move one change from issue to production without carrying every handoff yourself.
- LinearRead issuerequirements
- GitHubChange codepull request
- CICheck buildverified
- DeployReleaseproduction
- StatusReportclosed loop
Each step runs only after you expose and authorize that capability.
Smaller surface. Same completed work.
Code Mode cut the tool wall and token load without losing the result.
- 10/10
- tasks completed
- 72.0% fewer
- request + output tokens
- 7 vs 215
- initial choices
Run June 2026 with the real-world large MCP suite, openai-codex/gpt-5.5, 10 tasks, 2 runs per task, and a large no-fixture MCP stack.Review the benchmark
Get one real result.
Set up Caplets and run OSV. No authentication required.
npm install -g capletscaplets setupRead bootstrap skilldetect -> ask -> plan -> approve