Open source · MCP protocol · Coming soon

The runtime skill discovery
layer for agents.

100,000+ agent skills exist. Your agents have access to zero of them unless you've manually installed each one. We're building the runtime discovery layer.

Without Caplo
agent runtime
> Create a PDF report from this quarterly data
 
Generating PDF with reportlab...
Adding tables and charts...
✗ Tables overflow page margins
✗ Charts rendering as black boxes
✗ Font embedding missing — garbled text on Windows
 
Let me try a different approach...
With Caplo
agent runtime
> Create a PDF report from this quarterly data
 
✓ Loaded skill: pdf-report-generation
 
Building layout with proper page breaks...
Embedding fonts for cross-platform support...
Rendering charts as vector graphics...
Adding table of contents with page numbers...
 
✓ report-q4.pdf — 12 pages, ready to share
200+
Curated skills at launch
<1s
Search to load
0
Tokens at idle

Agents are only as good as the expertise they can access. Today, that means pre-installed skills manually found, manually vetted, manually configured. With 100,000+ skills in the ecosystem and no runtime discovery, most agent expertise goes unused. Caplo changes that. One connection, and your agents find what they need, when they need it.

One MCP connection. Every agent.

Claude Code logoClaude Code
Cursor logoCursor
Codex logoCodex
Gemini CLI logoGemini CLI
Windsurf logoWindsurf
Copilot logoCopilot
How it works

Your agent needs expertise. It asks. It gets it.

Discover

Agent encounters an unfamiliar task and queries for matching skills in natural language. Results in under 500ms.

Load

The right skill loads into context — instructions, examples, best practices. Only when needed, only what’s relevant.

Execute

Agent follows the loaded expertise and completes the task correctly. No restart. No config change.

Why Caplo

Built for trust, speed, and zero waste.

The index

Finding the right skill is harder than using it.

500+ skills sourced from official vendor collections and vetted community repositories. Every skill passes automated security screening — injection detection, payload analysis, exfiltration pattern matching. New skills added weekly.

Context budget

Zero tokens until activation.

Nothing loads at startup. Search results are lightweight. Full skills load on demand. The difference between five available skills and five hundred.

Security

36% of community skills have security flaws.

341 malicious skills found in a public registry. 36% of community skills have flaws. Caplo indexes from trusted sources only. Every skill is scanned before it's queryable.

Private indexes

Your org's expertise, queryable.

Deployment playbooks. Internal SDK patterns. Compliance procedures. Add private skills to a team index alongside the public one. Your agents discover both. Nobody outside your org sees your skills.

FAQ

Common questions

The runtime discovery layer
for agent skills.

Open source server. Hosted index free during early access.

caplo