How to Find and Evaluate MCP Servers on Smithery, Glama, and MCP.so

TL;DR - Key Takeaways

• MCP servers are plugins that connect AI assistants (Claude, Cursor, GPT) to external tools and data sources via the Model Context Protocol.
• Smithery (11,414+ servers) is the best choice for one-click install and hosted, auth-managed servers.
• Glama (47,981+ servers) offers the most rigorous quality grading (A–F across License, Quality, Maintenance) and the largest index - use it for deep evaluation.
• MCP.so is the cleanest discovery-first interface for browsing and exploring by category.
• 66% of MCP servers contain critical or blocker-level code issues (Queen's University, 2025) - quality signals matter more than you think.
• Always check: verification badge, usage count, GitHub stars, last-updated date, license, and README quality before installing anything.

---

What Are MCP Servers?

MCP stands for Model Context Protocol - an open standard that Anthropic introduced on November 25, 2024, to solve a real fragmentation problem in AI tooling. (For a full primer, see what MCP is.)

Before MCP, every AI framework (LangChain, LlamaIndex, AutoGen) had its own way of connecting to external tools. A Stripe integration built for LangChain wouldn't work in LlamaIndex. Developers kept rewriting the same connectors from scratch. MCP fixes that by providing a single, universal protocol.

An MCP server is a lightweight program that wraps an external service - a database, an API, a file system - and exposes it to any MCP-compatible AI client. Think of it as a plugin. Connect a Gmail MCP server to Claude, and Claude can read, draft, and send emails. Connect an Exa Search server to Cursor, and your coding assistant can search the live web.

Who uses MCP clients? Claude Desktop, Cursor, VS Code Copilot, ChatGPT, and dozens of open-source agents. As of May 2025, the MCP Python SDK was pulling ~1.8 million weekly downloads and the NPM package ~6.9 million - the ecosystem is moving fast.

---

Why Use a Directory to Find MCP Servers?

You could search GitHub directly. But with tens of thousands of MCP servers now in the wild, that's like searching npm without npmjs.com.

Directories solve three specific problems:

• Discovery: Filter by category, capability, or use case instead of guessing repo names.
• Quality signals: Directories surface usage counts, star ratings, verification badges, and code-quality grades that raw GitHub search doesn't show.
• Installation shortcuts: Platforms like Smithery let you go from "found it" to "running in Claude" in under a minute.

There's also a safety argument. A 2025 study from Queen's University analyzed 1,899 open-source MCP servers and found that 66% contain critical or blocker-level code smells and 14.4% have critical bugs. Beyond maintainability, 7.2% had general security vulnerabilities (including credential exposure) and 5.5% showed signs of tool poisoning - where a malicious server manipulates an AI into taking harmful actions. Installing a random server without checking its quality signals is a real risk. Good directories help you filter that noise.

The three main MCP server directories each take a different approach. (There's also the official MCP server registry, which standardizes how servers describe themselves.) Here's how to use each one.

---

Smithery - How to Find & Evaluate MCP Servers

Smithery (smithery.ai/servers) is the closest thing the MCP ecosystem has to Docker Hub. It's built around speed: from finding a server to running it in your AI client in under a minute.

What's in the Index?

Smithery indexes 11,414+ MCP servers. The front page shows featured and trending servers with live usage counts - a fast way to see what the community is actually running.

How to Search and Filter

• Keyword search: Type a tool name, capability, or service (e.g., "postgres", "slack", "web search").
• Verified filter: Narrow to servers that carry Smithery's official Verified badge - these have been reviewed by the Smithery team.
• Tag-based filtering: Browse by functional tags like "search", "productivity", "developer tools".
• Deployment status: Filter for hosted (remote) vs. locally-run servers.

Quality Signals on Smithery

Smithery doesn't use letter grades. Its two main trust signals are:

• ✅ Verified badge - Smithery's own review mark. Prioritize these.
• 📊 Usage count - Real adoption numbers. The Filesystem server has 45.56k uses, Gmail 39.20k, Exa Search 39.94k. High usage means the community has stress-tested it.

Each server card also links directly to its GitHub repository, so you can inspect the code, check the issue tracker, and verify the last commit date yourself.

Top Servers by Usage (June 2026)

Server	Usage Count
Google Sheets	45,560
AI Research Assistant (Semantic Scholar)	43,870
Exa Search	39,940
Gmail	39,200
Google Calendar	15,310
Mesh MCP	12,940
Brave Search	10,120
Slack	9,420

Smithery's Standout Features

One-click install is the headline. Hit "Add to toolbox" on any server card and Smithery walks you through connecting it to Claude, Cursor, or another MCP client - no manual JSON config editing required.

Smithery Connect handles OAuth (the industry-standard authorization flow that lets apps access your accounts securely without sharing passwords) automatically. For servers like Gmail or Google Sheets, Smithery manages the credential handshake for you.

Observability dashboard: Once a server is running, you can track call volume and errors - useful for debugging agent workflows.

CLI: Run npx smithery to install and manage servers from the terminal.

How to Submit a Server

Go to smithery.ai/servers/new and follow the submission form.

---

Glama - How to Find & Evaluate MCP Servers

Glama (glama.ai/mcp/servers) is the most comprehensive MCP server registry available. If Smithery is Docker Hub, Glama is the security-conscious enterprise registry. It's the right tool when you need to evaluate before you install.

What's in the Index?

As of June 24, 2026: 47,981 servers, 6,241 connectors, 291,305 individual tools. Updated daily.

How to Search and Filter

Glama's filtering is the most powerful of the three platforms:

• Sort by: Search Relevance, Recent Usage, Date Added, Date Updated, Weekly Downloads, GitHub Stars, or Recent GitHub Stars.
• 86 curated categories: Databases, Developer Tools, File Systems, Browser Automation, Version Control, Search, Communication, AI & ML, Coding Agents, RAG Systems, Code Execution, Observability, Marketing, E-commerce, and more.
• Tool-level search: This is unique to Glama. Instead of searching for a server by name, you can search by what it does - e.g., "query Postgres", "send email", "execute Python". Glama searches across 291,305 indexed tools and returns the servers that expose that specific capability.
• Official badge: Filters for servers submitted by the service's own development team.

Glama's A–F Quality Grading System

This is Glama's defining feature - and the most rigorous quality scoring in the MCP ecosystem. Every server gets three separate grades:

License Grade Is the license open-source friendly? An A means MIT or Apache 2.0. An F means the license is missing, proprietary, or problematic for most use cases.

Quality Grade Code quality, security posture, and adherence to best practices. Glama uses a hybrid pipeline: static analysis (automated code scanning, similar to how SonarQube works - a tool that checks code for bugs, vulnerabilities, and style issues without running it) combined with an LLM jury (Claude-3.7-sonnet, GPT-4o, and Gemini-2.5-pro each independently evaluate the code) plus manual review in isolated virtual machines. An A means high quality and secure. An F means critical failures.

Maintenance Grade Update frequency, community activity, and bug-handling responsiveness. An A means the project is highly active. An F means it's been abandoned or hasn't seen recent updates.

Why does this matter? Because the Queen's University study found that 66% of MCP servers have critical-level code issues. Glama's grading is the fastest way to filter out that 66% without reading every line of code yourself.

Real Examples from Glama (June 2026)

Server	License	Quality	Maintenance	Weekly Downloads	GitHub Stars
Filesystem MCP Server	A	A	B	263,182	87,454
DataNexus MCP	A	A	A	55	1,132 recent
SEO Review Tools	A	B	D	8	2
TorrentClaw-MCP	A	A	F	16	2

Notice the SEO Review Tools server: solid license, decent quality, but a D on maintenance (last updated June 2025 - a full year ago). The TorrentClaw server scores an F on maintenance despite passing quality checks. These grades save you from installing something that won't be patched when a bug surfaces.

Glama's Standout Features

• Tool-level search across 291,305 tools - find servers by capability, not just name.
• Largest index of the three platforms at 47,981 servers.
• Daily updates - grades reflect the current state of each repository.
• "Official" badge for provider-maintained servers.

---

MCP.so - How to Find & Evaluate MCP Servers

MCP.so (mcp.so) is the discovery-first option. It's community-driven, visually clean, and great for browsing when you're not sure exactly what you're looking for.

What's in the Index?

MCP.so claims one of the largest collections of MCP servers, with community submissions from across the ecosystem. It's integrated with the Awesome MCP Servers list on GitHub - a curated community repository that's become a go-to reference.

How to Search and Filter

• Keyword search across server names and descriptions.
• Category browsing: Browse by functional area - developer tools, productivity, data, search, and more.
• Tag filtering: Narrow by specific tags attached to each listing.
• Curated collections: The platform features hand-picked collections and highlighted servers, which is useful when you want a starting point rather than a blank search box.

Quality Signals on MCP.so

MCP.so is lighter on structured quality data than Glama or Smithery. The main signals are:

• GitHub stars - community endorsement at a glance.
• Community signals - featured status and inclusion in curated collections suggests some editorial curation.
• Integration with Awesome MCP Servers - servers that appear on this well-maintained GitHub list have at least passed community scrutiny.

MCP.so's Standout Features

• Clean, approachable UI - the lowest barrier to entry of the three platforms.
• Curated collections - useful for themed discovery (e.g., "best servers for coding agents").
• Community-driven - new servers appear quickly as the community submits them.

How to Submit a Server

Submissions are handled directly through the MCP.so website.

---

Side-by-Side Comparison Table

Feature	Smithery	Glama	MCP.so
Total Servers	11,414+	47,981+	Large collection
Quality Grading	Usage count + Verified badge	A–F grades (License, Quality, Maintenance)	GitHub stars
Categories	Tag-based	86 curated categories	Category browsing
Tool-level search	No	Yes (291,305 tools)	No
One-click install	Yes	No	No
Auth management	Yes (Smithery Connect)	No	No
Official badge	Yes (Verified)	Yes (Official)	No
Daily updates	Yes	Yes	Community-driven
Submit a server	smithery.ai/servers/new	glama.ai/mcp/servers	Via site
Best for	Quick install + hosted servers	Deep evaluation + research	Discovery + browsing

---

How to Evaluate Any MCP Server: Universal Checklist

Regardless of which MCP server registry you use, run through this checklist before adding any server to a production workflow.

1. Verification / Official Status

Is the server verified by the platform, or does it carry an "Official" badge from the provider itself? An official Stripe MCP server maintained by Stripe's own team is a fundamentally different risk profile from a community-built clone.

2. Usage or Download Count

Real adoption is the best proxy for real-world reliability. A server with 40,000 uses on Smithery or 263,000 weekly downloads on Glama has been stress-tested by thousands of users. A server with 3 downloads hasn't.

3. GitHub Stars

Stars are a community endorsement signal. They're not perfect - a server can have stars and still be abandoned - but combined with other signals they're useful. Check the star velocity too: recent stars (Glama shows this) indicate active interest.

4. Maintenance Grade / Last Updated Date

This is the one most people skip. An unmaintained server won't receive security patches. On Glama, look for an A or B Maintenance grade. On Smithery, click through to the GitHub repo and check the last commit date. If it hasn't been touched in 6+ months, think twice.

5. Code Quality

Glama's Quality grade is the fastest shortcut here. If you're evaluating a server not on Glama, pull up the GitHub repo and look for: test coverage, a CI/CD badge (continuous integration - automated testing on every code change), and an active issue tracker. A repo with 50 open unacknowledged issues is a warning sign.

6. License

Check that the license is compatible with your use case. MIT and Apache 2.0 are the most permissive and widely used. A missing license is a legal grey area - avoid it for anything production-facing. Glama's License grade makes this instant.

7. Documentation Quality

Does the README explain what the server does, how to configure it, and what permissions it requires? A server with a one-line README is harder to audit and harder to trust.

8. Security - Permissions and Auditability

What does the server actually need access to? A web search server that requests write access to your file system should raise flags. For sensitive integrations (email, databases, payment systems), read the source code or at least skim it. The Queen's University study found credential exposure as the most common vulnerability type - 3.6% of servers had it. (For a full vetting workflow, see our MCP server security checklist.)

For sensitive integrations, going one step further than a manual read is worth it — auditing MCP servers with mcp-scan catches prompt injection and tool poisoning that a quick skim won't.

Quick rule of thumb: If a server touches sensitive data (email, databases, payment APIs), spend 10 minutes on the GitHub repo before installing. For low-stakes tools (web search, documentation lookup), the platform's quality signals are usually enough.

---

Which Platform Should You Use?

The honest answer: use all three, for different jobs.

Use Smithery when:

• You want to get something running in Claude or Cursor right now.
• You need OAuth-managed auth for Google Workspace, Slack, or similar services.
• You're prototyping and speed matters more than exhaustive vetting.
• You want to monitor server usage with an observability dashboard.

Use Glama when:

• You're evaluating servers for a production or team workflow.
• You want to search by what a server can do, not just its name.
• You need to audit license, code quality, and maintenance status before committing.
• You're comparing multiple servers for the same task (e.g., three different Postgres MCP servers).

Use MCP.so when:

• You're new to MCP and want to explore what's possible.
• You're browsing by category without a specific server in mind.
• You want to see what the community is building and featuring right now.

The power move: Search Glama for quality-graded candidates, then install the winner via Smithery. Once you've shortlisted, our roundup of the best MCP servers in 2026 is a good sanity check against the most widely adopted options.

---

FAQ

What is an MCP server? An MCP server is a program that wraps an external service - a database, an API, a file system - and exposes it to AI assistants via the Model Context Protocol. Think of it as a plugin that gives your AI assistant new capabilities, like reading your email or querying a database.

What is the Model Context Protocol? The Model Context Protocol (MCP) is an open standard introduced by Anthropic on November 25, 2024. It provides a universal way for AI applications (like Claude, Cursor, and ChatGPT) to connect to external tools and data sources without custom integrations for each one.

Which MCP server directory has the most servers? Glama has the largest index with 47,981+ servers as of June 2026, followed by Smithery with 11,414+. MCP.so also hosts a large community-driven collection but doesn't publish an exact count.

Is Smithery free to use? Smithery offers free access to browse and install MCP servers. Some hosted server features and advanced auth management may have usage limits or paid tiers - check smithery.ai for current pricing.

How do I know if an MCP server is safe to install? Use Glama's A–F quality grades as a first filter. Then check: Is it verified or official? Does it have significant usage or download numbers? When was it last updated? Does the README explain what permissions it needs? For sensitive integrations, read the source code on GitHub before installing.

What does Glama's quality grade actually measure? Glama's Quality grade combines static code analysis (automated scanning for bugs, vulnerabilities, and code smells), an LLM jury (Claude-3.7-sonnet, GPT-4o, and Gemini-2.5-pro each evaluate the code independently), and manual review in isolated virtual machines. An A means high quality and secure; an F means critical failures were found.

What is "tool poisoning" in the context of MCP servers? Tool poisoning is an MCP-specific attack where a malicious server embeds hidden instructions in its tool descriptions, tricking an AI assistant into taking harmful actions - like exfiltrating credentials or writing malicious code. The Queen's University study found 5.5% of analyzed servers showed signs of tool poisoning. This is why installing only from verified or well-graded sources matters.

Can I submit my own MCP server to these directories? Yes. Submit to Smithery at smithery.ai/servers/new, to Glama via the "Add Server" button on glama.ai/mcp/servers, and to MCP.so directly through their website.

What's the difference between a "Verified" badge on Smithery and an "Official" badge on Glama? Smithery's Verified badge means Smithery's team has reviewed the server. Glama's Official badge means the server was submitted by the organization that owns the underlying service (e.g., the Stripe MCP server submitted by Stripe). Both are trust signals, but they measure different things.

---

Useful Sources

• Anthropic - Introducing the Model Context Protocol - Original announcement, November 25, 2024.
• modelcontextprotocol.io - Official MCP Specification - The authoritative protocol spec.
• arxiv.org - "MCP at First Glance: Studying the Security and Maintainability of MCP Servers" - Queen's University empirical study of 1,899 MCP servers (2025).
• smithery.ai/servers - Smithery MCP server directory.
• glama.ai/mcp/servers - Glama MCP server registry.
• mcp.so - MCP.so community directory.
• github.com/modelcontextprotocol/servers - Anthropic's official open-source MCP server repository.