Skip to content

klim agents

klim agents is klim’s view into the AI agent ecosystem on your machine. It cuts across multiple agent CLIs — currently Claude Code and GitHub Copilot CLI plus the MCP registry — and surfaces five entity types under a single set of subcommands:

TypeWhat it is
marketplacesSources an agent CLI pulls plugins, skills, and MCPs from.
pluginsFirst-party / community extensions installed for a provider.
skillsReusable prompts / playbooks loaded by an agent.
mcpsModel Context Protocol servers — tools an agent can call.
sessionsSaved or active conversations you can resume / inspect.

The same data also drives the TUI’s Agents tab — see the Agents Tab guide for the interactive view.

Terminal window
klim agents [command] [flags]

Running klim agents with no subcommand prints a summary of every detected entity, grouped by provider.

CommandWhat it does
klim agents listList discovered agent entities; filter by --type, --provider, --installed, --available, --search.
klim agents searchFuzzy search across all entities, or scope with type:query (e.g. plugin:auth).
klim agents launchOpen an agent session with a chosen skill / plugin / saved session.
klim agents marketplacesBrowse / add / remove marketplaces (provider-specific).
klim agents pluginsList / install / remove plugins.
klim agents skillsBrowse skills.
klim agents mcpsList / install / remove MCP servers.
klim agents sessionsList, resume, or delete agent sessions.
klim agents doctorDiagnose provider detection and cache freshness.
klim agents refreshInvalidate the agents scan cache and rescan from disk.

These flags apply to most subcommands; individual command pages note any extras.

FlagDescription
--providerLimit to one provider: claude-code, copilot-cli, or mcp-registry.
--typeFilter by entity type: `marketplace
--installedShow only installed entities (plugins / MCPs).
--availableShow only available (non-installed) catalog entries.
--searchSame as klim agents search <query>.
--refreshIgnore the cache and rescan from disk.
--outputtext (default), json, or yaml.
Terminal window
# Survey everything on this host
klim agents
# Just Claude Code's MCP servers
klim agents list --provider claude-code --type mcp
# Fuzzy search across everything
klim agents search react
# Scoped search — only plugins
klim agents search plugin:auth
# Launch a saved session
klim agents launch --session "claude:home%2Fuser%2Frepo"
# Pipe to jq / yq
klim agents list --output json | jq '.entities[] | select(.type=="plugin")'
klim agents list --output yaml > agents-snapshot.yaml

klim doesn’t speak to remote APIs to enumerate agent state — every provider has a well-known on-disk layout under the user profile, and klim parses those directly:

ProviderLayout
Claude Code~/.config/claude-code/ (Linux/macOS), %APPDATA%\Claude\ (Windows). Sessions, plugins, MCPs, marketplaces.
GitHub Copilot CLI~/.config/github-copilot/ and ~/.copilot/. Plugins, MCPs, session history.
MCP registryA bundled catalog of well-known MCP servers, refreshable via klim agents refresh.

The scan is cached per-host (~/.klim/cache/agents-cache.yaml) so repeated runs are fast; pass --refresh to force a rescan.