Caplets, capabilities instead of giant tool walls Skip to content
Caplets
svg]:opacity-70 hover:[&>svg]:opacity-100 focus-visible:ring-outline/50 transition-[color,box-shadow] outline-none focus-visible:ring-3" data-slot="sheet-close" aria-label="Close dialog">
Close
Give your agent Capabilities, not a tool wall
Turn sprawling MCP servers into focused capability cards. Your agent can start with one typed route, zoom into tools only when needed, and keep huge schemas out of the prompt until they matter.
First route the agent sees<br>One card opens into inspect, search, schema, and call only when needed.<br>osv / inspect / search_tools / get_tool / call_tool
svg]:px-4 [&_svg:not([class*='size-'])]:size-4.5" data-slot="button" href="#install"> Install & run setup svg]:px-4 [&_svg:not([class*='size-'])]:size-4.5" data-slot="button" href="https://docs.caplets.dev"> Read the docs<br>npm install -g caplets
Copy
Requires Node 24+. Run setup from the install section next.
*]:min-w-0"> Why Caplets
Agents do not need every tool at once.
Direct MCP flattens provider APIs into the prompt. The agent spends context reading tool names, giant schemas, and setup assumptions before it can do the work.
Caplets keeps the first surface small, then opens the exact operation path when the agent asks.
Direct MCP
With Caplets<br>Direct MCP<br>Too many tools<br>Every downstream operation lands in the agent's first view.
With Caplets<br>The agent starts with a named capability and opens only the route it needs.
Direct MCP<br>Too much schema<br>Large tool definitions compete with the user's actual task for context.
With Caplets<br>Schemas stay behind inspect, search, and get_tool until they matter.
Direct MCP<br>Too much setup<br>Every agent repeats provider wiring, OAuth, secrets, and MCP config.
With Caplets<br>One Caplets surface can be reused locally or from a remote server.
*]:min-w-0"> Setup
Start with the smallest useful Caplet.
caplets setup wires the agent integrations you choose. Add OSV first because it needs no auth; bring in GitHub or Sourcegraph after the discovery path feels right.
svg]:px-4 [&_svg:not([class*='size-'])]:size-4.5 mt-8" data-slot="button" href="https://github.com/spiritledsoftware/caplets/tree/main/caplets"><br>Explore more Caplets
discover cards<br>inspect tools<br>call typed
OSV GitHub Sourcegraph<br>img:first-child]:pt-0 *:[img:first-child]:rounded-t-xl *:[img:last-child]:rounded-b-xl gap-6 py-6 activation-card overflow-hidden" data-size="default" data-slot="card"> OSV<br>Add OSV<br>Vulnerability lookups by package, purl, commit, or batch query, with no auth ceremony.<br>Best first install: it proves the Caplet flow before OAuth or provider secrets enter the picture.<br>osv /inspect /search_tools /get_tool /call_tool<br>OSV is public. If the trial fails, check Node 24+ and rerun caplets setup.
OSV capability commands<br>01 caplets setup<br>02 caplets install spiritledsoftware/caplets osv<br>03 codex "try using the osv caplet"
img:first-child]:pt-0 *:[img:first-child]:rounded-t-xl *:[img:last-child]:rounded-b-xl gap-6 py-6 activation-card overflow-hidden" data-size="default" data-slot="card"> GitHub<br>Add GitHub<br>Repositories, issues, pull requests, branches, commits, and reviews behind one capability card.<br>This is the stress case: a valuable MCP server that is painful when every operation is exposed at once.<br>github /inspect /search_tools /get_tool /call_tool<br>If setup fails, check Node 24+, rerun caplets setup, and confirm GH_TOKEN.
GitHub capability commands<br>01 export GH_TOKEN=github_pat_...<br>02 caplets setup<br>03 caplets install spiritledsoftware/caplets github<br>04 codex "try using the github caplet"
img:first-child]:pt-0 *:[img:first-child]:rounded-t-xl *:[img:last-child]:rounded-b-xl gap-6 py-6 activation-card overflow-hidden" data-size="default" data-slot="card"> Sourcegraph<br>Add Sourcegraph<br>Hosted code search for examples, references, and implementation patterns across repositories.<br>Good for search-first work: the agent asks for matches, then opens only the operations it needs.<br>sourcegraph /inspect /search_tools /get_tool /call_tool<br>If setup fails, check Node 24+, rerun caplets setup, and finish Sourcegraph OAuth login.
Sourcegraph capability commands<br>01 caplets setup<br>02 caplets install spiritledsoftware/caplets sourcegraph<br>03 caplets auth login sourcegraph<br>04 codex "try using the sourcegraph caplet"
*]:min-w-0"> Benchmark proof<br>The result survived the surface cut.
Same task set, same 10/10 completion. Code Mode kept the agent focused by cutting the prompt surface instead of flattening every tool into view.
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.<br>The live run rows are listed here; the linked benchmark document covers the harness, deterministic surface check, and reproduction path.
svg]:px-4 [&_svg:not([class*='size-'])]:size-4.5 mt-8" data-slot="button" href="https://github.com/spiritledsoftware/caplets/blob/main/docs/benchmarks/coding-agent.md"><br>Read...