mcp

CLI that turns MCP servers into terminal commands.

mcp sentry search_issues '{"query": "is:unresolved level:error"}'

No SDKs. No wrapper code. One binary, any MCP server.

What it does

MCP (Model Context Protocol) servers expose tools — search issues, query logs, send messages, read files. This CLI lets you call those tools directly from your terminal. You configure a server once, then use it like any other command.

$ mcp slack list_channels
$ mcp grafana search_dashboards '{"query": "api-latency"}'
$ mcp sentry get_issue_details '{"issue_id": "12345"}'

Output is always JSON. Pipe it, jq it, script it.

Quick start

Install:

# Homebrew (macOS and Linux)
brew tap avelino/mcp https://github.com/avelino/mcp.git
brew install avelino/mcp/mcp

# Docker
docker pull ghcr.io/avelino/mcp
alias mcp='docker run --rm -v ~/.config/mcp:/root/.config/mcp ghcr.io/avelino/mcp'

# Pre-built binary from GitHub Releases
# Download the latest from https://github.com/avelino/mcp/releases

# From source (requires Rust)
cargo install --path .

Add a server from the registry:

mcp add filesystem

Or add an HTTP server manually:

mcp add --url https://mcp.sentry.dev/sse sentry

See what tools are available:

mcp sentry --list

Call a tool:

mcp sentry search_issues '{"query": "is:unresolved"}'

That's it. You're using MCP.

How it works

graph LR
    You["You<br/>(terminal)"] --> CLI["mcp CLI"]
    CLI <-->|"JSON-RPC 2.0<br/>stdio or HTTP"| Server["MCP Server"]
    Server <--> API["Service API"]
    Config[("servers.json")] -.-> CLI

    style CLI fill:#4a9,color:#fff

The CLI reads your config (~/.config/mcp/servers.json), connects to the server using stdio or HTTP, and speaks JSON-RPC 2.0 to call tools. Authentication (OAuth 2.0, API tokens) is handled automatically.

Configuration

Servers live in ~/.config/mcp/servers.json:

{
  "mcpServers": {
    "slack": {
      "command": "npx",
      "args": ["-y", "slack-mcp-server@latest", "--transport", "stdio"],
      "env": {
        "SLACK_MCP_XOXP_TOKEN": "${SLACK_TOKEN}",
        "SLACK_MCP_TEAM_ID": "${SLACK_TEAM_ID}"
      }
    },
    "sentry": {
      "url": "https://mcp.sentry.dev/sse"
    }
  }
}

Three types of servers:

• Stdio — the CLI spawns a local process (command + args)
• HTTP — the CLI connects to a remote URL (url + optional headers)
• CLI — wraps any command-line tool as an MCP server (command + cli: true)

Environment variables use ${VAR_NAME} syntax and are resolved at runtime.

Authentication

For HTTP servers that require auth, mcp handles it automatically:

1. OAuth 2.0 — If the server supports it, mcp opens your browser, completes the flow, and saves the token
2. Manual token — If OAuth isn't available, it guides you to get a token with service-specific instructions
3. Config headers — You can set Authorization headers directly in config

Tokens are saved in ~/.config/mcp/auth.json and refreshed automatically.

Commands

Piping JSON from stdin

If you don't pass JSON arguments on the command line, mcp reads from stdin:

echo '{"query": "is:unresolved"}' | mcp sentry search_issues

Docker

The CLI is available as a multi-arch Docker image (amd64/arm64):

# Run directly
docker run --rm -v ~/.config/mcp:/root/.config/mcp ghcr.io/avelino/mcp --list

# Call a tool
docker run --rm -v ~/.config/mcp:/root/.config/mcp ghcr.io/avelino/mcp sentry search_issues '{"query": "is:unresolved"}'

# Pass environment variables for servers that need them
docker run --rm \
  -v ~/.config/mcp:/root/.config/mcp \
  -e GITHUB_TOKEN \
  ghcr.io/avelino/mcp github search_repositories '{"query": "mcp"}'

# Use an alias for convenience
alias mcp='docker run --rm -v ~/.config/mcp:/root/.config/mcp ghcr.io/avelino/mcp'
mcp sentry --list

Available tags:

You can also run the proxy as a container instead of a background service. Your MCP client starts and stops it:

{
  "mcpServers": {
    "all": {
      "command": "docker",
      "args": [
        "run", "--rm", "-i",
        "-v", "/Users/you/.config/mcp:/root/.config/mcp",
        "ghcr.io/avelino/mcp:full", "serve"
      ]
    }
  }
}

A backend whose command is a binary from your host isn't in either image. mcp --list marks those unavailable with the reason, so run it in the container before switching a client over. See the Docker guide.

CLI as MCP

Any command-line tool becomes an MCP server — no code, no wrapper:

{
  "mcpServers": {
    "kubectl": {
      "command": "kubectl",
      "cli": true,
      "cli_only": ["get", "describe", "logs", "version"]
    }
  }
}

mcp parses --help automatically to discover subcommands and flags, then exposes them as MCP tools:

$ mcp kubectl --list
kubectl_get        Display one or many resources
kubectl_describe   Show details of a specific resource
kubectl_version    Print the client and server version information

$ mcp kubectl kubectl_get '{"args": "pods -A", "output": "json"}'

Works with kubectl, docker, terraform, git — anything with --help. See the CLI as MCP guide for details.

Proxy mode

Use mcp serve to expose all your configured servers as a single MCP endpoint. Configure it once, use it from any LLM tool:

{
  "mcpServers": {
    "all": {
      "command": "mcp",
      "args": ["serve"]
    }
  }
}

This works with Claude Code, Cursor, Windsurf, or any MCP-compatible client. Tools are namespaced as server__tool (e.g. sentry__search_issues). See the proxy mode guide for details.

Smart resource management

Every MCP client (Claude Code, Cursor, Windsurf) spawns all backend servers at startup and keeps them alive forever. With 10 servers and 3 sessions open, that's 30 idle processes eating ~3 GB of RAM.

The mcp proxy fixes this with persistent tool cache, lazy initialization, shared backends across clients, and adaptive idle shutdown:

• Instant startup — tools are cached to disk and served immediately, even before backends connect
• One backend = one process, no matter how many clients — 5 editor sessions hitting slack share a single slack-mcp-server child. Calls run in parallel via JSON-RPC id multiplexing on stdio.
• Backends only connect when you actually use them (background refresh keeps the cache fresh)
• Idle backends are shut down automatically (1-5 min based on usage frequency), with a warm-up grace period so a brand-new backend isn't reaped before its first use
• Tools stay visible — reconnection is transparent on next call
• Cache invalidates automatically when backend config changes
• Zero orphans: every spawned child is reaped on shutdown, panic, cancel, or stalled close (kill_on_drop everywhere)

{
  "mcpServers": {
    "slack": {
      "command": "npx",
      "args": ["@anthropic/mcp-slack"],
      "idle_timeout": "adaptive"
    },
    "sentry": {
      "url": "https://mcp.sentry.io",
      "idle_timeout": "never"
    }
  }
}

Read the full story: MCP servers are draining your hardware

Audit logging

Every operation is logged — tool calls, searches, config changes, proxy requests. Query the log with filters or stream it in real-time:

mcp logs                          # recent entries
mcp logs --server sentry --errors # sentry failures only
mcp logs --since 1h               # last hour
mcp logs -f                       # follow mode (tail -f)
mcp logs --json | jq '...'        # pipe to jq

Logs are stored locally in an embedded database. No external services, no network calls. See the audit logging guide for configuration and details.

Environment variables

Documentation

Full documentation: docs/

Development

cargo build
cargo test

Contributing

1. Fork the repo
2. Create your branch (git checkout -b my-feature)
3. Make your changes
4. Run tests (cargo test)
5. Submit a pull request

License

MIT