Smithery.ai – MCP Tools Marketplace

TL;DR

TL;DR: Smithery.ai is a marketplace and registry for MCP (Model Context Protocol) servers, letting you browse, connect, and deploy tools for AI agents in minutes — with auth and credentials managed for you.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: smithery.ai — verified
• HN launch thread: news.ycombinator.com/item?id=48593013 — verified via HN Algolia
• Source last checked: 2026-07-31

What Is Smithery.ai?

Smithery.ai is an MCP server marketplace. The Model Context Protocol (MCP) is an open standard that lets AI agents use external tools — similar to how a plugin system works for a web browser. Smithery acts as the registry and distribution layer: you browse available MCP servers, connect them to your AI client, and the platform handles auth, credentials, and session management so you do not have to.

From the site:

Connect agents to thousands of tools and services. Auth, credentials, and sessions handled for you.

The platform lists hundreds of MCP servers covering domains like web search, finance, weather, AI news, documentation lookup, and more. Each server is a discrete tool that an AI agent can call at runtime.

What MCP Is (and Why It Matters)

MCP (Model Context Protocol) is a protocol that lets AI models call external tools during inference. Rather than hard-coding tool integrations into a model, MCP defines a standard interface: an AI client that supports MCP can load any compatible server and call its tools. Think of it as a universal plugin standard for AI agents.

Smithery is the package registry for this ecosystem — npm for MCP servers, if you like that analogy.

How Smithery Works

Browsing and Selecting Tools

Visit smithery.ai to browse available MCP servers. Each listing shows what the tool does, how many times it has been used, and what credentials or API keys it requires. For example:

• Exa Search (12.76k uses) — Fast web search and crawling for AI agents
• Context7 (5.42k uses) — Fetch up-to-date, version-specific library documentation
• Agent News (22.94k uses) — AI news with citations and confidence scores
• Wayforth (26.91k uses) — API runtime for AI agents across 4,974 verified APIs

Connecting to an MCP Client

Once you pick a server, Smithery provides the configuration to add it to your MCP-compatible AI client (Claude Desktop, Cursor, Kiro, etc.). The platform supports two connection models:

1. Smithery-hosted — Smithery runs the MCP server and handles credentials. You connect via an API key or Smithery session token.
2. Self-hosted — You deploy the MCP server yourself (open-source on GitHub) and register it with Smithery for discovery.

Auth and Credentials

The key value-add: Smithery manages the auth layer. For servers that require API keys (e.g., Exa Search, Wayforth), Smithery can hold those credentials on your behalf, so you do not expose keys in your local MCP config. This is similar to how a password manager works, but for AI tool credentials.

Pricing

Smithery uses a pay-as-you-go model. From the site:

Pay per query — no subscription required.

Specific per-tool pricing varies. Some servers have free tiers; others charge per call. The Wayforth MCP, for example, charges $0.002 USDC per query via the x402 payment protocol.

Practical Evaluation Checklist

• Browse smithery.ai and find a tool relevant to your workflow
• Verify the tool works with your specific MCP client
• Check whether the server is Smithery-hosted or self-hosted
• Test the auth flow — confirm credentials are handled by Smithery, not stored locally
• Evaluate the free tier vs. paid usage costs for your use case
• Check the HN launch thread for community feedback on reliability

Security Notes

• Smithery holds credentials on your behalf — this means you are trusting Smithery’s infrastructure with API keys for third-party services
• Review what data Smithery logs about your tool usage
• For high-security use cases, prefer self-hosted MCP servers where you control the credentials entirely
• MIT-compatible open-source servers are available for most popular tools on GitHub

FAQ

Q: Do I need to use Smithery to use MCP? A: No. MCP is an open standard — you can find MCP servers on GitHub, run them yourself, and configure them manually. Smithery is a convenience layer: a registry with one-click connect, auth management, and usage tracking.

Q: Is Smithery free to use? A: Some MCP servers have free tiers. Others charge per query. The platform uses pay-as-you-go pricing with no subscription. Specific costs depend on which servers you use.

Q: Can I self-host an MCP server instead of using Smithery’s hosted version? A: Yes. Most servers listed on Smithery are open-source. You can deploy them yourself and skip Smithery’s managed auth layer entirely.

Q: Which AI clients support MCP? A: Claude Desktop (Anthropic), Cursor, Kiro, and any client that implements the MCP specification. Support varies by client — check each tool’s documentation for compatibility details.

Q: Is there a limit on how many MCP servers I can connect? A: No hard limit. You can connect as many servers as your MCP client supports loading simultaneously.

Conclusion

Smithery.ai fills the gap between “I want my AI agent to do X” and “which tool do I install, how do I configure it, and where do I put the API key?” By managing auth and providing a one-click connect flow, it lowers the friction of building MCP-based AI workflows.

The marketplace model makes it easy to discover new MCP servers — the hardest part is not finding tools, it is knowing what exists. If you are running MCP-compatible AI clients and want a managed experience with credential handling, Smithery is worth evaluating.