llm-mcp-client Brings MCP Tools to Simon Willison's LLM CLI
Simon Willison released llm-mcp-client 0.1a0 on July 31, an alpha plugin that exposes tools from remote Model Context Protocol servers to his LLM command-line utility. The plugin discovers a server's tools through an MCP toolbox, supports stateless and legacy transport modes, and can prefix tool names when multiple servers might collide; it remains an initial prototype rather than a core LLM feature.
Simon Willison released llm-mcp-client 0.1a0 on July 31, adding an installable Model Context Protocol client to his open-source LLM command-line tool. The project is an alpha plugin and its GitHub release labels it an initial prototype.
What the plugin does
After installation, the plugin registers an MCP toolbox. A user supplies the URL of an MCP server, and the toolbox discovers the tools exposed by that server and makes them available to an LLM prompt or chat session. The documented command-line flow uses LLM's -T option to add the toolbox and --td to display tool-call details while they execute.
The plugin can also be used from Python. It supports the modern stateless MCP transport as well as the older session-based mode, with an automatic setting by default. When a prompt uses more than one MCP server, callers can add prefixes to prevent overlapping tool names from colliding.
Willison introduced the package in a broader post about the July 28, 2026 MCP specification. He said the simpler stateless protocol made this iteration practical after earlier attempts, and demonstrated the client against a Datasette MCP endpoint that exposes read-only database tools. That example shows the integration path; it is not evidence that arbitrary MCP servers are safe or compatible.
Why the alpha matters
LLM already supports model providers and locally registered tools. This plugin adds a standard remote discovery path, so developers can test an MCP server from the same CLI without writing a one-off adapter for every tool. The explicit stateless and legacy modes also make the transport choice visible during the protocol transition.
The package is not yet part of LLM core. Willison described direct integration as a possibility only after the plugin is more mature. Practitioners evaluating the alpha still need to review each server's permissions, data exposure and tool behavior before making those capabilities available to a model.
Key Points
- 1llm-mcp-client 0.1a0 registers an MCP toolbox that discovers server tools and exposes them to LLM prompts and chats.
- 2The alpha supports stateless, legacy and automatic transport selection, plus prefixes for avoiding tool-name collisions across servers.
- 3The release is an installable plugin and initial prototype; it has not been folded into LLM core.
- 4A retrieved Datasette example demonstrates the integration path but does not establish the safety of arbitrary MCP servers.
Scoring Rationale
A useful early integration for developers who use Simon Willison's LLM CLI and want standards-based remote tool discovery. The release is still an initial alpha prototype with no independent adoption or production evidence.
Sources
Primary source and supporting public references used for this report.
Practice with real Logistics & Shipping data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Logistics & Shipping problems
