CVMI (wip)

ContextVM Interface (CVMI) is a CLI tool that allows you to navigate and use the ContextVM protocol. It provides a set of tools and skills to help you interact, and implement the protocol.

Note: This project is a fork of the skills CLI by Vercel Labs.

Quick Start

# Install ContextVM skills interactively
npx cvmi add

# Install a specific skill from the ContextVM repository
npx cvmi add --skill overview

Roadmap

• cvmi add - Install skills with interactive picker
• cvmi add --skill <name> - Install specific skills
• cvmi serve - Expose a server (gateway)
• cvmi use - Use a server from nostr as stdio (proxy)
• cvmi cn - Compile a server to code (ctxcn)
• cvmi call - Call methods from a server
• cvmi inspect - Inspect server schema

Configuration

Configuration is stored in JSON format with the following priority:

1. CLI flags (highest priority)
2. Custom config: --config <path>
3. Project-level: ./.cvmi.json
4. Global: ~/.cvmi/config.json
5. Environment variables

Global config path: ~/.cvmi/config.json (separate from ~/.agents/ used for skills)

Nostr MCP environment variables:

• CVMI_SERVE_* / legacy CVMI_GATEWAY_* for serve/gateway settings
• CVMI_USE_* / legacy CVMI_PROXY_* for use/proxy settings

Additional serve env vars:

• CVMI_SERVE_URL / CVMI_GATEWAY_URL to set the remote Streamable HTTP MCP server URL

Logging environment variables (SDK-level): The underlying @contextvm/sdk uses these env vars to control logging:

Variable	Values	Description
LOG_LEVEL	debug, info, warn, error, silent	Minimum log level to output (default: info)
LOG_DESTINATION	stderr, stdout, file	Where to write logs (default: stderr)
LOG_FILE	path string	Path to log file (used when LOG_DESTINATION=file)
LOG_ENABLED	true, false	Disable all logging with false (default: true)

Example: Run serve with debug logging to a file

LOG_LEVEL=debug LOG_DESTINATION=file LOG_FILE=./cvmi.log cvmi serve -- npx -y @modelcontextprotocol/server-filesystem /tmp

Example: Run use with only warnings and errors

LOG_LEVEL=warn cvmi use npub1q...

Example global config (~/.cvmi/config.json):

{
  "serve": {
    "url": "https://my.mcp.com/mcp",
    "command": "npx",
    "args": ["@modelcontextprotocol/server-filesystem", "."],
    "privateKey": "nsec1...",
    "relays": ["wss://relay.damus.io"],
    "public": false,
    "encryption": "optional"
  },
  "use": {
    "privateKey": "nsec1...",
    "relays": ["wss://relay.damus.io"],
    "serverPubkey": "npub1...",
    "encryption": "optional"
  }
}

Note: For serve, you should configure either serve.url (remote Streamable HTTP MCP server) or serve.command/serve.args (spawn local stdio MCP server).

About quoting commands

cvmi serve spawns the MCP server directly (no shell). Prefer passing the command and its arguments as separate tokens:

cvmi serve -- npx -y @modelcontextprotocol/server-filesystem /tmp

If you accidentally pass a full command as a single quoted string (e.g. "npx -y ..."), cvmi will split it into an executable + args for you.

Passing environment variables to the spawned MCP server

Use --env / -e (repeatable):

cvmi serve -e LOG_LEVEL=debug -- npx -y @modelcontextprotocol/server-filesystem /tmp

You can also set it in config under serve.env.

Note: The CLI auto-generates a private key if none is provided. Keys can be specified in hex format (with or without 0x prefix) or NIP-19 bech32 format (nsec1... for private keys, npub1... for public keys).

License

MIT