Smithery CLI Skill

Agent skill for discovering, configuring, and invoking MCP servers and tools directly through the Smithery command line.

The Smithery CLI Skill is a specialized operational skill developed by Smithery and Arcade AI for autonomous agents. It equips language models such as Claude Code, Codex, or Cursor with canonical procedural guidelines and command syntax to search for MCP servers, register endpoints, inspect tool hierarchies, and invoke tools directly via the smithery command-line utility. The primary open-source repository is maintained at github.com/arcadeai-labs/smithery-cli (previously smithery-ai/cli), with the skill definition located at skills/smithery-ai-cli/SKILL.md.

Capabilities Imparted to the Agent

While many coding and automation assistants have built-in support for MCP concepts, they often lack runtime knowledge on how to dynamically discover and attach integrations on the fly. This skill bridges that capability gap. Whenever a user task calls for an external service, such as managing GitHub issues, querying Jira backlogs, pulling database records, or checking observability metrics, the agent loads the skill and executes a deterministic workflow.

First, the agent uses smithery mcp search to identify qualified servers matching the required integration or service domain. Next, the server is attached via smithery mcp add under a defined local identifier or URL.

For targeted inspection, the agent uses smithery tool list to explore tool groups hierarchically, for example drilling down into github issues, keeping token overhead low by avoiding massive bulk schema dumps. Before invoking mutations, smithery tool get verifies the exact JSON schema for input parameters. Finally, smithery tool call dispatches the tool call using validated payloads and returns the output to the active agent context.

Architectural Boundaries Between Skill and CLI Binary

This catalog record represents the instructional skill, not the executable binary or an isolated server daemon. To leverage the workflows defined here, the host system must have the smithery executable available via global npm installation. For authentication against secured services, the skill directs the agent to initiate browser-based confirmation via smithery auth login rather than attempting automated credential guessing.

Security Controls and Permission Boundaries

Because dynamic tool discovery can grant an agent mutating powers across external third-party infrastructure, such as deleting database rows or merging pull requests, security safeguards are strictly codified. Agents are instructed never to execute state-changing actions without explicit human approval. Furthermore, local authentication credentials and access tokens generated by the CLI must remain strictly within protected local credential vaults and never surface in chat histories or public repositories.

Installation and Commands

Installation works with npm install -g smithery@latest and requires Node.js 20 or newer. The commands are grouped into mcp (search, add, list, remove, publish), tool (list, find, get, call), auth (login, logout, whoami, token) and namespace (list, use). Servers can be published by URL or as an .mcpb bundle with smithery mcp publish. The project is licensed under AGPL-3.0.

Licensing and Catalog Status

The repository is distributed under the GNU Affero General Public License version 3 (AGPL-3.0). On September 9, 2026, the GitHub repository recorded 833 stars. This count represents repository-wide adoption rather than a sub-path metric. The skill and CLI are freely available for open-source workflows. Individual remote endpoints hosted via third-party registries may operate under their own terms of service, which is why Skill Road does not cite fixed pricing in this public entry.

Installation instructions

What the Smithery CLI skill is

Smithery is a registry service and tool catalog for the Model Context Protocol standard, or MCP, which lets AI agents access external tools and data sources in a standardized way. The associated Smithery CLI, developed in the arcadeai-labs/smithery-cli repository, is described by the provider as the command-line tool that lets users connect their agents to thousands of skills and MCP servers directly from the terminal. Instead of configuring each integration by hand, the CLI bundles search, installation, management, and invocation of MCP tools into a single command-line program. The skill is aimed primarily at developers who want to wire up several AI agents to different external services without maintaining a separate configuration file for every connection by hand.

Step 1: Installation and core functionality

According to the official README, installation happens via the Node package manager npm using the command npm install global smithery latest, and Node.js version 20 or newer is required. After installation, several command groups become available. The MCP server area lets users search the Smithery registry for suitable servers, add a found server by its URL, list existing connections, and remove connections that are no longer needed. A second command group revolves around the actual tools provided by connected MCP servers: users can list available tools, search them by keyword or intent, inspect details for a single tool, and finally invoke it with matching parameters.

Step 2: Authentication, namespaces, and skill management

For accessing protected areas of the registry and publishing one's own server, the CLI includes an authentication section that supports browser-based OAuth login, logout, identity verification, and minting service tokens with optionally restricted permissions. In addition, namespace commands let users manage and switch between multiple workspaces, which is particularly useful for teams that maintain separate environments for development and production. Besides MCP servers, Smithery also supports its own skills registry, containing separate, reusable capability modules for agents, which is browsed and installed via an upstream installer called npx skills add.

Step 3: Practical use: connecting and publishing servers

Day-to-day usage typically starts with a search, for example for a GitHub tool, followed by adding the found server under a chosen identifier. Users can then search tools on the connected server and invoke them directly with structured arguments, for instance to create a new issue in a repository. Anyone running their own MCP server can publish it to the registry via the publish command, either directly by URL or as a bundled MCP bundle archive, making it discoverable to other users. For contributing to the CLI itself, the source code is openly available and can be cloned via git, built with the pnpm package manager, and tested locally.

Security, best practices, and limitations

Because the CLI manages access tokens for third-party services and grants agents access to potentially sensitive tools, authentication should always go through the intended OAuth and token mechanisms rather than being replaced by manually copying credentials into configuration files. Restricted service tokens with a defined policy are preferable when an agent should only have limited rights, instead of granting it full account access. A known risk factor in MCP integrations in general is that connected servers can perform actions on behalf of the agent, which is why only trusted servers from official sources should be connected. The limitation of the Smithery CLI is that it functions purely as a connection and management tool: it neither guarantees the quality of individual registered servers nor takes responsibility for their behavior, so reviewing individual tools before production use remains advisable.

Free
Provider
Smithery
License
AGPL-3.0
Last reviewed
09.09.2026

Repository and documentation

Categories

Compatible with

Claude Code Codex Cursor