Claude Code MCP uses an open standard to connect Claude Code with external tools and contextual data through local or remote servers, with explicit configuration, status checks, and permission controls.
What MCP Means in Claude Code
The Model Context Protocol (MCP) is an open standard for connecting AI agents to external systems. Its purpose is to provide a reusable connection model instead of requiring a separate custom integration for every combination of agent and system.
MCP uses a host-client-server architecture. The host is the AI application coordinating one or more clients, each client maintains a connection to an MCP server, and the server supplies capabilities or context that the host can use. The protocol’s data layer is based on JSON-RPC and supports capability and version discovery, notifications, and primitives including tools, resources, and prompts.
Tools and resources have different roles. Tools are executable functions that an AI application can invoke for operations such as API calls, file actions, or database queries. Resources provide contextual information such as file contents, database records, or API responses.
In this setting, Claude Code is the agentic coding tool acting as the application that uses those connections. Anthropic documents Claude Code as able to read codebases, edit files, run commands, and integrate with development tools across terminal, IDE, desktop, and browser surfaces.

How Claude Code Connects to MCP Servers
Claude Code documents four MCP server configurations: remote HTTP, remote SSE, local stdio, and remote WebSocket. The right configuration depends on whether Claude Code connects to a network endpoint or starts a process on the local machine.
For a remote HTTP server, the documented command pattern is claude mcp add --transport http <name> <url>. This identifies the transport explicitly and associates a chosen server name with its URL. An imported URL configuration without a transport type can fail because Claude Code interprets an entry with no type as a stdio server.
For a local stdio server, Claude Code launches a command and communicates with the resulting process. Its documented pattern is claude mcp add [options] <name> -- <command> [args...]. The separator distinguishes Claude Code options and the server name from the command and arguments used to start the local server.
Treat connection type as a functional setup choice: use a documented remote transport for a URL-based endpoint and stdio for a local launch command. This avoids a common mismatch before permissions or server behavior are considered.

Server Status, Permissions, and Practical Limits
Adding a configuration does not guarantee that its server is ready for use. Claude Code provides claude mcp list to list configured servers, claude mcp get <name> to inspect one, and claude mcp remove <name> to remove one. Inside Claude Code, /mcp displays server status.
A project-scoped server declared in .mcp.json may remain pending until it is reviewed and approved in an interactive Claude Code session. A server can also be rejected, disabled for the project, or blocked by an organization’s managed policy. These states make the status view an important part of troubleshooting.
Access should be reviewed separately from connectivity. Claude Code supports allow, ask, and deny permission rules. Its documented wildcard syntax includes mcp__*, which can deny all MCP tools. As a practical recommendation, review these rules before permitting an MCP tool to run.
There is also an efficiency tradeoff. Anthropic identifies two scaling concerns when agents connect to many MCP tools: tool definitions occupy context, while intermediate tool results pass through the model. Larger collections can therefore increase context use, latency, token consumption, and cost. Keeping only relevant, trusted connections available is a practical response to those documented limits.
Conclusion
MCP is useful when Claude Code needs a standardized connection to external actions or contextual data instead of a separate integration for each system. Begin with a trusted server, choose the remote or local transport that matches how it operates, and add it using the corresponding documented command. Then check claude mcp list and /mcp, inspect any pending, rejected, disabled, or policy-blocked state, and review the allow, ask, or deny rules before permitting its tools to run.
Disclosures and limitations
- This article was created with AI assistance and is based on official Anthropic, Claude Code, and Model Context Protocol documentation supplied for this entry.
- No hands-on product testing or personal use is claimed.
Sources
- Connect Claude Code to tools via MCP – Claude Code Docs — Claude Code Docs
- Code execution with MCP for efficient agents — www.anthropic.com
- Architecture overview – Model Context Protocol — Model Context Protocol
- Configure permissions – Claude Code Docs — Claude Code Docs
- Curated research source from code.claude.com — code.claude.com
Improve this guide
Report a problem or suggest the next question
Submissions enter a private editorial queue. They are reviewed and deduplicated before any public update or new page is created.
