The Model Context Protocol (MCP) has quietly become the standard way to give AI coding assistants superpowers. Instead of being limited to reading and writing files, MCP lets tools like Claude Code, Cursor, and Codex connect to databases, browsers, APIs, and more.
But with dozens of MCP servers out there, which ones are actually worth setting up? Here are 10 that I use regularly and recommend to every developer working with AI tools.
1. Filesystem
What it does: Gives your AI assistant controlled read/write access to files and directories outside the default workspace.
Why it matters: Essential when your AI needs to access config files, logs, or assets in other directories.
claude mcp add filesystem -- npx @anthropic/mcp-filesystem ~/projects ~/configs
2. GitHub
What it does: Search repos, read issues, create PRs, manage branches — all through your AI assistant.
Why it matters: You can say "check my open PRs" or "create an issue for this bug" without leaving your terminal.
claude mcp add github -- npx @anthropic/mcp-github```
{% endraw %}
Requires a GitHub personal access token set as {% raw %}`GITHUB_TOKEN`{% endraw %}.
## 3. Brave Search
**What it does**: Web search through the Brave Search API — your AI can look things up in real time.
**Why it matters**: When your AI needs current documentation, error messages, or API references that aren't in its training data.
{% raw %}
```bash
claude mcp add brave-search -- npx @anthropic/mcp-brave-search```
Requires a free API key from [brave.com/search/api](https://brave.com/search/api/).
## 4. SQLite
**What it does**: Query and modify SQLite databases directly from your AI session.
**Why it matters**: Perfect for prototyping, exploring local databases, or building data-driven features. Your AI can write and run SQL queries, inspect schemas, and analyze results.
```bash
claude mcp add sqlite -- npx @anthropic/mcp-sqlite --db-path ./mydb.sqlite
5. TokRepo
What it does: Search, browse, and install from a registry of 200+ curated AI assets — skills, MCP configs, prompts, and workflows.
Why it matters: Instead of googling for "best Claude Code skill for code review" or digging through GitHub, your AI can search TokRepo directly and install assets in one step.
claude mcp add tokrepo -- npx tokrepo-mcp-server
Once connected, you get tools like tokrepo_search, tokrepo_trending, and tokrepo_install. Say "find me a good git workflow" and it returns matching assets from the registry with install commands. No API key needed.
6. Puppeteer
What it does: Browser automation — navigate pages, take screenshots, click elements, fill forms.
Why it matters: Great for testing web apps, scraping data, or debugging UI issues. Your AI can literally look at your app and tell you what's wrong.
claude mcp add puppeteer -- npx @anthropic/mcp-puppeteer```
{% endraw %}
## 7. Memory
**What it does**: Persistent key-value memory that survives across sessions using a local knowledge graph.
**Why it matters**: Your AI can remember project decisions, coding conventions, and context between conversations. No more re-explaining your architecture every session.
{% raw %}
```bash
claude mcp add memory -- npx @anthropic/mcp-memory```
{% endraw %}
## 8. Fetch
**What it does**: Makes HTTP requests — GET, POST, PUT, DELETE — to any URL.
**Why it matters**: Your AI can test APIs, pull data from endpoints, check if services are running, or interact with webhooks.
{% raw %}
```bash
claude mcp add fetch -- npx @anthropic/mcp-fetch```
{% endraw %}
## 9. Sequential Thinking
**What it does**: Provides a structured thinking framework that helps your AI break down complex problems step by step.
**Why it matters**: For architectural decisions, debugging complex issues, or planning multi-step implementations, this server helps the AI reason more carefully instead of jumping to solutions.
{% raw %}
```bash
claude mcp add sequential-thinking -- npx @anthropic/mcp-sequential-thinking```
{% endraw %}
## 10. PostgreSQL
**What it does**: Connect to PostgreSQL databases — run queries, inspect schemas, analyze data.
**Why it matters**: If your production or staging database is PostgreSQL, this lets your AI understand your data model, write queries, and help debug data issues.
{% raw %}
```bash
claude mcp add postgres -- npx @anthropic/mcp-postgres postgresql://user:pass@localhost/mydb```
{% endraw %}
---## How to Set Up Multiple MCP Servers
You can add MCP servers one by one with {% raw %}`claude mcp add`{% endraw %}, or configure them all at once in your {% raw %}`.mcp.json`{% endraw %}:
{% raw %}
```json
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": ["@anthropic/mcp-filesystem", "~/projects"]
},
"github": {
"command": "npx",
"args": ["@anthropic/mcp-github"] },
"tokrepo": {
"command": "npx",
"args": ["tokrepo-mcp-server"]
}
}
}
Start with 2-3 that match your workflow, then add more as needed. Each server runs as a lightweight subprocess — they don't consume resources until your AI actually calls them.
Wrapping Up
MCP servers turn your AI assistant from a code-completion tool into something closer to a full development partner. The ones listed here cover the most common needs: file access, search, databases, browsers, APIs, and asset management.
What MCP servers are you using? I'm always looking for new ones to try — drop your favorites in the comments.
Top comments (0)