klim agent

klim agent 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:

Type	What it is
marketplaces	Sources an agent CLI pulls plugins, skills, and MCPs from.
plugins	First-party / community extensions installed for a provider.
skills	Reusable prompts / playbooks loaded by an agent.
mcps	Model Context Protocol servers — tools an agent can call.
sessions	Saved 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.

Usage

klim agent [command] [flags]

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

Subcommands

Command	What it does
`klim agent list`	List discovered agent entities; filter by `--type`, `--provider`, `--installed`, `--available`, `--search`.
`klim agent search`	Fuzzy search across all entities, or scope with `type:query` (e.g. `plugin:auth`).
`klim agent launch`	Open an agent session with a chosen skill / plugin / saved session.
`klim agent marketplace`	Browse / add / remove marketplaces (provider-specific).
`klim agent plugin`	List / install / remove plugins.
`klim agent skill`	Browse skills.
`klim agent mcp`	List / install / remove MCP servers.
`klim agent session`	List, resume, or delete agent sessions.
`klim agent doctor`	Diagnose provider detection and cache freshness.
`klim agent refresh`	Invalidate the agents scan cache and rescan from disk.

Global flags

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

Flag	Description
`--provider`	Limit to one provider: `claude-code`, `copilot-cli`, or `mcp-registry`.
`--type`	Filter by entity type: `marketplace
`--installed`	Show only installed entities (plugins / MCPs).
`--available`	Show only available (non-installed) catalog entries.
`--search`	Same as `klim agent search <query>`.
`--refresh`	Ignore the cache and rescan from disk.
`--output`	`text` (default), `json`, or `yaml`.

Examples

# Survey everything on this host
klim agent

# Just Claude Code's MCP servers
klim agent list --provider claude-code --type mcp

# Fuzzy search across everything
klim agent search react

# Scoped search — only plugins
klim agent search plugin:auth

# Launch a saved session
klim agent launch --session "claude:home%2Fuser%2Frepo"

# Pipe to jq / yq
klim agent list --output json | jq '.entities[] | select(.type=="plugin")'
klim agent list --output yaml > agents-snapshot.yaml

Where klim looks

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:

Provider	Layout
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 registry	A bundled catalog of well-known MCP servers, refreshable via `klim agent refresh`.

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