A directory service that acts as the entry point for connecting AI agents to external tools and services. You can search for servers that support the Model Context Protocol (MCP) as well as Agent Skills, and connect them directly to your own agent or AI client. Web search, database queries, browser control, and business tool integrations can be pulled in as ready-made connections instead of being implemented one by one. The platform also takes care of the tedious parts of connecting — OAuth authentication, storage of credentials such as API keys, and session maintenance. Launched in December 2024 out of the United States, it also serves as a place where developers can publish and monetize their own MCP servers.
Key Features
- A directory of MCP servers and Agent Skills: A registry that lets you browse and search a large number of integrations. You can look for ready-made servers by purpose — web search, browser control, documentation lookup — and use them as they are
- Connections without OAuth setup: The platform advertises “no redirect URIs, client IDs, or secrets to configure,” and provides a hosted setup flow when a connection is created. Tokens are refreshed automatically
- Secure credential storage: Sensitive values such as API keys are stored encrypted and treated as write-only (they cannot be read back once saved). For browser and mobile use, short-lived service tokens with narrowed scope can be issued
- Three access paths: CLI, REST API, and SDK: A CLI that adds a connection with a single command, a REST API for managing connections programmatically, and a typed TypeScript SDK (
@smithery/api) — you can pick whichever suits the job - An aggregated URL per namespace: Integrations linked to your namespace can be exposed together as a single URL, usable from clients such as Claude, Cursor, ChatGPT, and MCP Inspector without reconfiguring each app
- Visible connection state and session management: Connections carry states such as connected, authentication required, input required, and error, and can hold metadata for associating them with users — which helps when an app serves many users
- Publishing and monetization for developers: You can publish your own MCP server to the registry so others can discover it, and there is a path to monetizing it
Pricing
| Plan | Price | What it covers |
|---|---|---|
| Free tier | $0 | Try the basics — browsing and searching the registry, creating connections |
| Paid plans | Check the site | Higher usage limits and additional features. See the official pricing page for amounts |
Pricing is as of August 2026. Specific amounts for paid plans may not render depending on how the page is loaded, so please check the official pricing page directly for the latest and most accurate terms.
Pros & Cons
✅ Pros
- Finding and connecting MCP servers happens in one place, cutting the work of writing connection code for each one
- Parts that are quietly heavy to implement — OAuth setup and token refresh — can be left to the platform
- The aggregated namespace URL lets you register several tools with an AI client as a single endpoint
- With a CLI, REST API, and SDK all available, you can move from hands-on experimentation to production integration on the same mechanism
- Because credentials are stored write-only, your app does not have to hold keys in plain text
⚠️ Cons
- The structure of the paid plans is hard to read, so estimating cost requires checking the official page or contacting the company
- It assumes an understanding of MCP itself, so there is an initial hurdle without experience in AI agent development
- Since it mediates connections to external services, it adds one more place that holds your credentials. For highly confidential work, its handling policy should be reviewed
- Many of the servers in the registry are third-party, so quality and maintenance vary from server to server
- Japanese-language documentation and write-ups are still scarce
Comparison with Similar Services
| Criterion | Smithery | Official MCP registry | Composio | Zapier MCP |
|---|---|---|---|---|
| Main role | Search plus hosted connections for MCP servers and skills | An official listing of MCP servers | An integration layer of tools for agents | Existing automation services offered over MCP |
| Auth handling | Automates OAuth and token refresh | Not provided (listing is the point) | Provides credential management | Follows each service’s own account |
| Connection methods | CLI / REST API / SDK / aggregated URL | Follow each server’s instructions | Mainly SDK and API | A generated MCP URL |
| Developer publishing | Publishing and monetization supported | Publishing supported | Adding integrations supported | Not applicable |
| Best suited for | Developers bundling several MCPs | People looking for servers of known provenance | Developers wanting deep SaaS integration | No-code automation users |
Who Is It For
- Developers who want to add web search, browser control, or business tool integration to their own AI agent or chatbot quickly
- People juggling several MCP servers whose per-client configuration has become unwieldy
- Small product teams who would rather not own OAuth implementation and token refresh themselves
- Engineers learning MCP who first want an overview of what servers exist
- Developers who want to publish their own MCP server and build a path to getting it used
Summary
Smithery is a directory platform that supports MCP servers and Agent Skills across three fronts: finding them, connecting them, and managing them. Its handling of OAuth and credentials in particular clears away work that tends to bite later in agent development. On the other hand, the terms of the paid plans need to be confirmed on the official page, and the scope of the credentials you hand over is worth sorting out before adoption. A good start is to browse the registry on the free tier and connect a single server you actually want to use.