AgentLayer▸docs
Connect it

Packs

A pack connects your agent to one outside system, such as GitHub, a database or Slack. Each one ships with the plugin and stays off until you switch it on.

A pack is a group of tools that reach one outside system: your pull requests, a database, a browser, a support queue. All ten ship inside the plugin, so there is nothing to download. A pack is off until you switch it on. Until then its tools ask before every call, and a tool whose key is missing names the key it needs.

The ten packs

PackWhat it gives youWhat it needs
GitHubPull requests, issues, CI runs and their logs, everything on GitHub waiting on you, and checkouts kept current by syncThe gh CLI and a token per repo owner
DatabaseRead-only Postgres queries and schema, plus a scratch copy of a local databaseA connection string per database
BrowserScreenshots, PDFs, pages as Markdown, recordings, scripted flows and web searchNothing. A Perplexity key for web search is optional
SEOSearch Console, PageSpeed, live Google results, domain authority and WordPress contentA Google OAuth client and your site's address. SerpAPI and Open Page Rank keys are optional
SlackA read-only pulse across your channels, the replies you owe and the threads waiting on your teamA Slack app in your workspace, created once, and your own sign-in
HelpdeskSupport tickets from SupportCandy or Pylon, replies drafted as private notes, bugs raised where you track themA SupportCandy application password or a Pylon API token
App StoreAny app's App Store reviewsNothing
XcodeBuilding, testing, running and debugging Apple apps and Swift packagesmacOS with Xcode 27 or later, and three steps only you can run
APIAPI requests drafted as files for Bruno, Postman or curlNothing. A client is optional
cmuxThe AgentLayer sidebar in cmux: every running session, and your agent's projects, reports and goalsmacOS with cmux. No keys, tools or permissions

Every account a pack can use, and what it costs, is on Accounts and costs. Most are free.

Switch one on

Setup offers the packs near its end. Any time after, run the configure-skills skill and choose Configure a skill pack. You can tick several at once.

  1. Answer the pack's questions. Each pack asks only what it needs: which databases, which helpdesk, which API client.
  2. Add the keys in your editor. The skill tells you which lines to add and to which file (where keys live). It never asks for a key in chat, because a pasted key would end up in the session transcript.
  3. Relaunch. New keys and permissions load at the next session, so exit your host and start it again from the home.

Switching a pack on adds its tools to permissions.allow in the home's .claude/settings.json, so they stop asking before each call. A pack you did not pick is not listed there, which keeps that file an accurate record of what the agent can reach. MCP tools shows which tools are granted when.

Switch one off

Run configure-skills and choose Deconfigure a skill pack. It removes the pack's permissions and offers to remove its keys. The pack's skills stay, since they are part of the plugin, but their tools go back to asking.

Other MCP servers and skill libraries

configure-skills also connects things the plugin does not ship:

  • A remote MCP server you already use, such as a log platform's hosted one. The skill registers it in <HOME>/.mcp.json with its token kept in the secrets store, and the server refuses to start while the token is missing.
  • Third-party skill libraries from skills.sh. They install into <HOME>/.claude/skills/ as links, so re-running the install picks up the author's updates. A third-party skill runs in your session with your permissions and nobody vets what it does, so read it before you install it. Skills lists the two libraries the skill offers.

Packs tuned to one industry are on the site's modules page.

On this page