MCP: Connect Claude Code to Everything
Model Context Protocol is the most important AI coding extension in 2026 — one standard to connect AI with all your dev tools
What is MCP?
MCP (Model Context Protocol) is an open standard by Anthropic that lets AI assistants securely connect to external data sources and tools. With MCP, Claude Code can directly query databases, access GitHub, search docs, and control browsers without manual copy-paste.
AI Application
Applications driven by models such as Claude that connect to external tools and data sources over MCP. Claude Code, this guide's subject, reads your codebase, edits files, runs commands, and integrates with your development tools.
Tools and Data Sources
External capabilities delivered through MCP Servers. Servers provide functionality through three building blocks: tools the model can invoke, readable resources, and structured prompt templates.
MCP Core Concepts: Host, Client and Server
MCP divides the ecosystem into hosts, clients and servers: the host application orchestrates access, each client keeps a dedicated connection to one server, and servers turn external data and tools into protocol capabilities.
MCP Host
The AI application you use directly: it coordinates and manages one or multiple MCP clients and their access.
MCP Client
The protocol component the host creates for each server: it keeps a dedicated connection and turns app needs into requests and responses.
MCP Server
An independent program that provides context to clients and exposes external capabilities through the primitives tools, resources and prompts.
Popular MCP Servers
The MCP ecosystem already spans files, repositories, databases, browsers, error monitoring, and team collaboration. Beyond the reference servers maintained by the MCP steering group (checked 2026-09-22), community and vendor servers keep growing; a few common ones are listed below.
Filesystem Server
Read/write local files, search and manage directory structures
GitHub Server
Manage repos, PRs, issues, and search code
Database Server
Query databases directly, analyze schemas, generate SQL
Browser Server
Control browsers for automated testing and web interaction
Sentry Error Monitoring
View error logs, analyze stack traces, auto-locate bugs
Slack Messaging
Send messages, search channel history, manage workflow notifications
Configure MCP in Claude Code
Simply add MCP Server configuration to your settings.json
The stdio transport: local server processes
With stdio, the client launches the server as a subprocess and exchanges messages over standard input and output. It suits tools that need direct access to your machine: just provide the launch command and its arguments, like the npx filesystem example.
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": [
"-y",
"@modelcontextprotocol/server-filesystem",
"/path/to/project"
]
}
}
}The SSE transport for remote servers
SSE connects clients to remote servers over HTTP. Claude Code docs mark it deprecated and recommend the HTTP transport where available; still, some services expose only an SSE endpoint, which is what this example configures (checked 2026-09-22).
{
"mcpServers": {
"remote-server": {
"type": "sse",
"url": "https://mcp.example.com/sse",
"headers": {
"Authorization": "Bearer your-token"
}
}
}
}MCP Security Best Practices
MCP Servers have powerful system access — proper security configuration is critical
Least Privilege
Grant only the minimum permissions needed — avoid root or admin accounts
Verify Server Sources
Only use official or trusted community-maintained Servers — review source code and dependencies
Network Isolation
Restrict Server network access — prevent connections to untrusted external endpoints
Audit Logging
Enable MCP call logging and regularly review tool invocation records for anomalies
Frequently Asked Questions
Do I have to write code to use an existing MCP server?
Using an existing server needs no code: the reference servers maintained in the official repository start with an npx command added to your configuration. You only write a server when you want to expose your own systems through the SDK.
Which file holds MCP server config?
Local and user entries live in .claude.json in your home directory; project entries go in .mcp.json at the repository root, where version control can share them. claude mcp add writes local scope by default, so choose another one with --scope. Checked against the official docs on 2026-09-22.
HTTP or SSE for a remote MCP server?
Choose HTTP for new remote connections: the docs list HTTP servers as the recommended way to reach remote MCP servers. SSE is marked deprecated on the same page, so keep it for services that publish nothing else. Checked 2026-09-22.
What to check when a server won't connect
Run claude mcp list first: Connected means the server is up, Needs authentication is a missing sign-in, Failed to connect is the actual failure. Then claude mcp get with the server name, or /mcp inside a session, shows the detail. If your model side goes through a gateway, note that MCP tool search is off by default and needs ENABLE_TOOL_SEARCH set explicitly. Checked 2026-09-22.
Is MCP only for Claude Code?
No. MCP is an open standard, and server entries are not tied to one client. The docs spell out how to bring across a server written for Claude Desktop, Cursor or any other MCP client: take its URL, its launch command, or the JSON block. Checked 2026-09-22.
Tool Categories You Can Connect
By purpose, common servers fall into the four categories below. The real power of MCP emerges when multiple servers work together, combining their specialized capabilities through a unified interface.
Developer and Code Tools
Connect Git and code hosting platforms to read, search, and manipulate repositories, bringing commits, branches, and reviews into the conversation.
Data and Storage
Perform secure file operations with configurable access controls, manage directories, and connect to databases to query data and inspect schemas.
Collaboration and Messaging
Integrate Slack workspaces: list channels, post messages, reply to threads, add reactions, and read channel history to keep the team in sync.
Web and Information Services
Servers such as Fetch handle web content fetching and conversion for efficient LLM usage; similar servers are commonly used for search and documentation lookup.
Use MCP via QCode.cc
MCP needs no different hookup: same key, same endpoint, and the mcpServers block just lives in your project
Tool side and model side stay separate
MCP servers go into Claude Code's own configuration: local and user entries in .claude.json, project entries in .mcp.json. Where your model requests go is a separate setting, so the two never overwrite each other. Checked 2026-09-22.
Your model credentials stay
Adding an MCP server leaves the key and endpoint QCode issued untouched. A server keeps its own secrets in its env or headers fields, and the docs note that --env takes several KEY=value pairs at once. The two credential kinds never replace each other. Checked 2026-09-22.
Share servers through the repo
A team can run on one server list: commit it to .mcp.json at the repository root. For safety, Claude Code asks you to approve each project server before using it in an interactive session, and claude mcp reset-project-choices clears those approvals. Checked 2026-09-22.
QCode also powers OpenAI Codex / GPT-5.6
Your QCode quota works across Claude Code and OpenAI Codex CLI — one shared balance, zero duplicate spend.
Start Using MCP with Claude Code
Enjoy the full power of the MCP ecosystem at low cost
Updated 2026-09-22
Create a QCode account
One key across Claude, GPT and Chinese models, billed per token — live rates on /models.