mcp-cli

Establishes connections to MCP servers through four distinct transport mechanisms (configuration file, direct command execution, HTTP, and Server-Sent Events) using the @modelcontextprotocol/sdk as the underlying protocol handler. The CLI abstracts transport selection logic, allowing users to connect via the same command interface regardless of whether the server is local, remote, or running as a subprocess, with automatic protocol negotiation and session management handled transparently.

Queries connected MCP servers to discover and list all available primitives (resources, tools, and prompts) using the MCP SDK's discovery APIs, then presents them in a formatted, interactive CLI menu with colored output and progress indicators. The discovery process automatically introspects server capabilities and populates a selectable list that users can navigate to choose which primitive to interact with, with metadata (descriptions, input schemas) displayed inline.

Wraps the @modelcontextprotocol/sdk to provide a compliant MCP client implementation that handles protocol details transparently. The CLI abstracts away MCP protocol specifics (message serialization, request-response matching, error handling) by delegating to the SDK, ensuring compatibility with any MCP server that implements the protocol specification. This abstraction allows users to interact with MCP servers without understanding the underlying protocol mechanics, while maintaining full protocol compliance.

Reads static resources (data, metadata, files) exposed by MCP servers by calling the server's resource read endpoint with a specified resource URI. The CLI handles resource selection from the discovered list, passes the URI to the MCP SDK's resource read method, and displays the returned content with appropriate formatting (text, JSON, or raw output depending on content type). Supports streaming large resources and handles errors gracefully with user-friendly messages.

Enables users to call MCP server tools by selecting from discovered tools, then interactively prompts for required and optional arguments based on the tool's JSON Schema input specification. The CLI uses the prompts library to collect user input, validates arguments against the schema, and passes them to the MCP SDK's tool call method. Results are displayed with formatted output, and errors are caught and presented with helpful context about what went wrong (e.g., missing required arguments, type mismatches).

Invokes MCP server prompts (template-based content generators) by selecting from discovered prompts, collecting user-provided arguments interactively based on the prompt's argument specification, and passing them to the MCP SDK's prompt call method. The CLI handles argument substitution into the prompt template and displays the generated response. Supports prompts with zero or multiple arguments, with validation ensuring required arguments are provided before invocation.

Reads and parses MCP server configuration from a file (in Claude Desktop format) that specifies server definitions with their command, arguments, and environment variables. The CLI loads this configuration, allows users to select which server to connect to, and establishes a connection by spawning the server process as a subprocess with stdio transport. This approach mirrors Claude Desktop's configuration model, enabling users to manage multiple server definitions in a single file and switch between them via CLI selection.

Spawns MCP servers directly from shell commands specified on the CLI (e.g., `mcp-cli exec 'node server.js'`), establishing a stdio-based transport connection to the spawned process. The CLI handles process lifecycle management (spawning, cleanup), stdio stream handling for MCP protocol messages, and error handling if the server process exits unexpectedly. This approach enables testing and using MCP servers without pre-configuration, useful for ad-hoc server invocation or development workflows.

Connects to MCP servers running remotely and exposed via HTTP or Server-Sent Events (SSE) transports by accepting HTTP/SSE endpoint URLs on the CLI. The CLI uses the eventsource library for SSE connections and native HTTP for request-response communication, establishing a remote transport session with the server. This enables interaction with MCP servers running on different machines or in cloud environments without requiring local process spawning, useful for accessing shared or managed MCP server instances.

Renders CLI output with syntax highlighting and colors using the yoctocolors library, and provides visual progress feedback using yocto-spinner for long-running operations (server discovery, tool execution, resource reading). The CLI automatically detects terminal color support and falls back to plain text if unsupported. Progress spinners display operation status (e.g., 'Discovering tools...') and are replaced with results or errors upon completion, improving user experience for interactive workflows.

Parses CLI arguments using the meow library and collects user input interactively using the prompts library. The CLI supports both command-line flags (e.g., `--server myserver`) and interactive prompts for missing or optional arguments. When users don't provide required information on the command line, the CLI prompts them to select or enter values interactively, with validation and error messages guiding them through the process. This hybrid approach supports both scripted and interactive usage patterns.