FOR DEVELOPERS 👋
You don't use sift9. Your coding agent does.
Right now, when your agent plans a task, it guesses which
tools to use — or you go find them yourself. sift9 closes
that gap. The agent sends its plan, and sift9 returns the
skills, plugins, MCP servers, hooks and rules that make the
job faster and more reliable. 200,000+ curated assets,
always updated, delivered in one call.
No account. No key. Install once, forget it's there.
INSTALL 🚀
Native plugin — Claude Code 2.1.289+:
curl -fsSL https://sift9.xyz/install | sh -s -- --plugin
Installs the routing skill, MCP connection and native activity panel.
Saving the first plan draft triggers search during planning, before
approval. Claude can load relevant skill guidance and refine the plan.
Outside plan mode, the skill guides discovery while forming a plan.
The hook never mounts or executes matches automatically. Plan edits
do not repeat search; Claude can search again when needs change.
/sift9-status shows activity and an unload button. /sift9-cleanup
releases mounts created here when no tools or agents are running.
Cleanup verifies the gateway session and retains pre-existing mounts.
Normal turn endings retain mounts for ongoing work; the skill guides
Claude to release each capability when done. Interruptions, API errors
and session endings attempt cleanup. Hard crashes rely on gateway
session expiry. Unmounting does not erase previously read skill text.
Requires Python 3.8+ for setup. Start a new session after installation.
Rerun with --plugin to update. Enable auto-update for the sift9
marketplace in /plugin. Migration removes the old managed reminder,
display and unchanged skill after plugin installation succeeds.
Edited skills are preserved and reported before migration.
The panel works in the terminal and Desktop Code tab. Other surfaces
can use the status command's text output. A plugin update may require
/reload-plugins or a new session.
Legacy setup for older Claude versions:
curl -fsSL https://sift9.xyz/install | sh
Installs the sift9 skill, MCP connection and activity display.
If Claude Code is missing, prints "Claude Code not found" and stops.
Requires Python 3.8+ for the local activity display. Run the installer
again to upgrade an existing setup; your hooks and status line stay.
Start a new Claude Code session and use /mcp to check the connection.
The skill tells Claude when to search and how to mount and release
capabilities through the MCP server. Describe your task normally;
/sift9 is optional. A prompt hook reminds Claude to search when the
plan needs extra capabilities, avoid repeating unchanged searches,
and release mounts after use. To invoke the skill explicitly:
/sift9 Extract tables from PDFs and export a validated spreadsheet
The reminder is guidance, not a guarantee of search on every task.
It runs locally, makes no network calls, and does not store your prompt.
The Claude Code footer shows actual sift9 activity:
sift9 · Finding suitable tools…
sift9 · Loading PDF extraction…
sift9 · PDF extraction loaded · 1 loaded
sift9 · Unloading PDF extraction…
sift9 · PDF extraction unloaded
Failed calls show a failure, not a successful load or unload.
"Ready" means the display is waiting for activity. Check /mcp for
connection status. The display tracks the main Claude Code session;
it does not trigger searches or unload tools by itself.
MCP endpoint: https://sift9.xyz/mcp
WITHOUT SIFT9 ❌
😤 Your agent picks the wrong MCP server for the job.
🔁 It writes code from scratch that a skill already solves.
🗑️ It wastes tokens re-discovering tools it should know about.
📚 You collect skills yourself — now the agent's context is
full of assets it doesn't need for this task.
🔍 Even in the age of AI, you still end up searching Google
for the right plugin before you start building.
WITH SIFT9 ✅
Install once. Everything else happens under the hood.
🎯 Your agent plans. sift9 finds the right assets for that
specific plan — ranked, curated, scoped to the task.
📦 Skills, plugins, MCP servers, subagents, hooks, rules,
workflow templates, and permission policies.
🌍 200,000+ assets from across the digital world, always
updated.
🙌 You do nothing. The agent calls sift9, gets back a
manifest, picks what it needs, and continues.
HOW IT WORKS ⚙️
1️⃣ Your agent has a plan.
2️⃣ It calls sift9 with the plan.
3️⃣ sift9 returns a ranked manifest of matching assets.
4️⃣ The agent picks what it needs.
5️⃣ The agent installs and continues building.
🔒 sift9 suggests skills, plug-ins, MCP and other
artifacts. Nothing runs server-side.