CLImax

Every CLI tool you want an AI agent to use needs a custom MCP server — bespoke code doing the same thing every time: parse arguments, run a subprocess, return output.

CLImax eliminates that boilerplate. Describe your CLI's commands in a YAML config, and CLImax serves them as MCP tools. If your CLI has --help, CLImax can expose it.

You need three things: CLImax, a YAML config describing your CLI's commands, and the CLI itself on PATH. Optionally, a policy file controls which tools are enabled, constrains argument values, overrides descriptions, and can route execution through Docker containers.

Exposing every subcommand as an individual MCP tool doesn't scale. A CLI like kubectl can have dozens of subcommands — combine a few CLIs and you're looking at 100+ tool definitions, each consuming context tokens just to be described. That overhead leaves less room for the actual task.

CLImax solves this with a two-tool pattern inspired by Cloudflare's Code Mode architecture:

The agent discovers capabilities on-demand. Two tool definitions, constant overhead, regardless of how many CLIs you wire up. For simpler setups, --classic mode registers all tools directly instead.

The easiest way to create a config is to let an LLM generate it. CLImax ships with a skill that teaches agents how to read --help output and produce valid YAML:

climax skill --install
# Then ask Claude: "Create a CLImax config for kubectl"

The agent inspects subcommands and generates a ready-to-use config. No manual YAML writing required.

Configs define what tools exist. Policies control what's allowed — and the two are intentionally separate. Policies can restrict which tools are enabled, constrain argument values, override descriptions, and route execution through Docker containers for sandboxing.

This means you can share a comprehensive git.yaml across your team while locking down individual environments: read-only tools in CI, restricted argument values per contributor, or fully containerized execution. The config stays the same; the policy shapes what any given deployment can actually do.

# readonly-git.policy.yaml
default: disabled
tools:
  git_status: {}
  git_log:
    args:
      max_count:
        max: 100
  git_diff: {}

Same git config, but the agent can only read — never write.

CLImax ships with configs for common tools:

# Install
uv tool install climax-mcp

# Run with a bundled config
climax git

# Combine multiple CLIs
climax claude git docker

# Classic mode — register all tools directly
climax --classic git

# List bundled configs
climax list

# List tools in a config
climax list git

# Validate a config
climax validate git