← Volver a las guías

What Is MCP? A 10-Minute Quick Start for Developers

What Is the Model Context Protocol?

The Model Context Protocol (MCP) is an open standard that lets AI applications connect to external tools, data sources, and reusable workflows. Instead of building a different integration for every assistant, developers can expose capabilities through an MCP server that compatible applications understand.

For example, a coding assistant could use one MCP server to read project documentation and another to query a database. MCP standardizes how the application discovers and accesses those capabilities. It does not make the model inherently more knowledgeable or grant unrestricted access to your computer.

The Architecture in One Minute

• Host: The AI application, such as Claude Code, that manages the user experience, model interaction, and permissions.
• Client: A component inside the host that maintains a connection to an MCP server.
• Server: A local process or remote service exposing capabilities through MCP.

MCP messages use JSON-RPC 2.0. During initialization, the client and server negotiate the protocol version and capabilities. Local servers commonly communicate over standard input and output, called stdio. Remote servers can use Streamable HTTP.

Servers can expose three core primitives: tools for actions such as searching or querying, resources for addressable context such as documents, and prompts for reusable interaction templates. Support varies by host; not every application exposes every primitive.

Quick Start: Connect Claude Code to Local Files

This example uses the filesystem reference server. You need Claude Code installed and a current Node.js LTS installation with npm and npx available.

• 1. Prepare a safe directory. Create a folder containing a few non-sensitive documents. Use its absolute path, such as /Users/alex/mcp-demo. Do not start with your home directory or a folder containing credentials.
• 2. Register the server. Run this terminal command, replacing the final path: claude mcp add --transport stdio local-files -- npx -y @modelcontextprotocol/server-filesystem /Users/alex/mcp-demo
• 3. Check the connection. Run claude mcp list, then open Claude Code and use /mcp to inspect the configured server and available connection controls.
• 4. Try a narrow request. Ask: “Use local-files to list the documents in /Users/alex/mcp-demo, then summarize the README. Do not modify any files.” Review any tool approval before accepting.

The filesystem server restricts operations to its allowed directories, but it includes write-capable tools: directory restrictions are not read-only permissions. For repeatable deployments, review the package and pin a tested version rather than always downloading the latest release.

What Happens When You Ask a Question?

The host discovers tools through tools/list. When the model proposes a tool invocation, the host applies its permission policy and can send tools/call to the server. The result returns to the application, which can include it in the model’s next turn.

This distinction matters: the model does not directly execute arbitrary server code. The application mediates execution. Also, tool results can contain malicious instructions, so treat retrieved content as untrusted data rather than authority to override your policies.

Where 59API Fits: Model Access, Not Tool Access

MCP connects your assistant to capabilities; a model API supplies inference. These are separate layers. Changing your model provider does not automatically install MCP servers, and an OpenAI-compatible endpoint does not automatically implement MCP.

59API provides low-cost, pay-as-you-go access to native Claude and GPT models without downgrading them, with compatibility for Claude Code, Codex, and OpenAI SDK clients. Its API origin is https://api.59api.com; follow its client-specific setup instructions for endpoint paths and authentication.

This can help control inference spending while you experiment with multi-step tool workflows. MCP server hosting, third-party services, and additional model turns may still add costs. If you want an economical starting point, consider signing up for 59API; a referral rebate is also available.

Before Connecting Production Systems

Use least-privilege credentials, keep secrets out of tool output, require approval for destructive actions, and log executions without leaking sensitive data. Start with one server and one measurable task. Add more capabilities only when they improve the workflow.

Precios · Docs