How to Add Agent Capabilities to Claude Code with MCP

Give Claude Code image generation, video, web search, and cloud storage through MCP. Step-by-step MCP server configuration guide plus the AnyCap one-command fast path.

by AnyCap

AI agent connecting to multiple capability servers through MCP protocol

Claude Code is a powerful terminal-based coding agent — but out of the box, it only edits files, runs shell commands, and manages git. It can't generate images, search the web, create videos, or store files in the cloud. MCP (Model Context Protocol) changes that.

MCP is the open standard that lets AI agents connect to external tools. Claude Code supports MCP natively. Configure a server, and Claude can call it directly — no copy-pasting, no browser tabs, no context switching.

This guide covers two paths: the manual MCP configuration (for individual servers) and the AnyCap fast path (one CLI for every capability your agent needs). If you haven't installed Claude Code yet, start with our Claude Code installation guide.

What Claude Code Can't Do (And MCP Fixes)

Capability Without MCP With MCP
Generate images Ask Claude, copy prompt to another tool Claude calls the image gen server directly
Search the web Open a browser, search manually Claude queries live search APIs
Create videos Use separate video tools Claude generates video from text or images
Cloud storage Local files only Claude uploads, shares, and retrieves from cloud
Publish content Manual deployment Claude deploys via publishing MCP
Query databases Manual SQL Claude queries Postgres, Supabase, etc.
GitHub automation Use gh CLI manually Claude creates PRs, reviews code, manages issues

Every "without MCP" column represents a workflow interruption. Every "with MCP" column is Claude doing the work while you stay focused.

Path 1: Manual MCP Configuration

How MCP Works

MCP servers expose tools through a JSON-RPC interface. Claude connects to them via two transport methods:

  • stdio — Local process communication. The server runs on your machine. Claude Desktop uses this primarily.
  • HTTP/SSE — Remote server communication. Claude Code and Claude Web support this for cloud-hosted servers.

Adding Servers to Claude Code

Interactive (stdio servers):

# GitHub server
claude mcp add github -- npx -y @modelcontextprotocol/server-github

# Brave Search
claude mcp add brave-search -- npx -y @modelcontextprotocol/server-brave-search

# Postgres
claude mcp add postgres -- npx -y @modelcontextprotocol/server-postgres

Interactive (HTTP servers):

claude mcp add --transport http my-server https://my-mcp-server.com/mcp

Project-level configuration (.mcp.json):

For team-wide consistency, define MCP servers in .mcp.json at your project root:

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_TOKEN": "ghp_your_token_here"
      }
    },
    "postgres": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres"],
      "env": {
        "DATABASE_URL": "postgresql://localhost/mydb"
      }
    }
  }
}

Commit this file — your whole team gets the same MCP servers automatically.

Adding Servers to Claude Desktop

Claude Desktop reads from claude_desktop_config.json:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": { "GITHUB_TOKEN": "ghp_your_token" }
    }
  }
}

Save, fully quit Claude Desktop (not just close the window), and restart.

Managing Servers

# List all configured servers
claude mcp list

# Inspect a specific server's tools
claude mcp get github

# Remove a server
claude mcp remove github

Configuring individual MCP servers works, but it means managing separate API keys, tokens, and config files for every capability. Image generation needs one server. Web search needs another. Video generation needs a third. Cloud storage, a fourth.

AnyCap gives you all of them through a single command:

npx -y skills add anycap-ai/anycap -a claude-code

What You Get

Capability MCP Server What Claude Can Now Do
Image Generation Nano Banana 2, Seedream 4.5 Create images from text, edit existing images, generate hero banners, product mockups
Video Generation Veo 3.1, Seedance 1.5 Pro Produce videos from text descriptions or still images
Web Search Brave, Exa, Firecrawl Search the live internet, crawl websites, extract structured data
Cloud Storage AnyCap Drive Upload files, generate share links, retrieve stored assets
Web Publishing AnyCap Page Deploy content as hosted web pages

One CLI. One authentication flow. Every capability. Run anycap login once, and every MCP server is authenticated. For a comparison of Claude Code with other coding tools, see our Claude Code vs Cursor comparison.

How It Works Under the Hood

AnyCap is an agent capability runtime. When you install it into Claude Code:

  1. The AnyCap CLI registers as an MCP client — Claude Code discovers available tools automatically
  2. Each capability maps to a production MCP server — Image generation, video, search, and storage each run on dedicated infrastructure
  3. Authentication is unified — One anycap login provisions access across all servers
  4. Tools appear in Claude's context — Ask Claude to generate an image or search the web, and it invokes the right server

Example: End-to-End Workflow

You're building a landing page. Without MCP, you'd switch between Claude Code, a separate image tool, a browser for research, and a file-sharing service. With AnyCap:

You: "Generate a hero image for a SaaS landing page with a dark theme and gradient accents"

Claude: [calls AnyCap image generation MCP server]
       Here's the generated hero image.

You: "Search for the latest pricing of competing SaaS tools"

Claude: [calls AnyCap web search MCP server]
       Based on current data: Tool A charges $29/mo, Tool B $49/mo.

You: "Upload the hero image to cloud storage and give me a share link"

Claude: [calls AnyCap Drive MCP server]
       Uploaded. Share link: https://drive.anycap.ai/abc123

One conversation. No tool switching. No copy-pasting.

Managing MCP Token Costs

Every MCP server you connect exposes tool definitions to Claude's context window. These consume tokens — invisibly, but significantly:

Servers Tools Context Used Impact
3–5 servers ~15–25 tools 2–4% Minimal
8–10 servers ~40–60 tools 8–15% Noticeable
15+ servers ~80+ tools 15–30% Significant — may reduce session quality

Best practice: Start with the capabilities you use daily. AnyCap's design keeps tool definitions lean — one unified tool surface instead of fragmented tool lists across servers. For full token management strategies, see our Claude Code rate limits guide.

Troubleshooting MCP Configuration

Problem Likely Cause Fix
Server not appearing Config file not loaded Fully quit and restart the host application
"Connection closed" Node.js version mismatch Use nvm: nvm install 22 && nvm use 22
Tools show as "disabled" Tool name conflict across servers Check for duplicate/conflicting tool names
Auth fails API key or token expired Verify credentials; regenerate if needed
First call hangs Reverse proxy timeout Increase proxy_read_timeout; disable SSE buffering

Run claude doctor before spending time on manual debugging — it auto-detects most configuration issues.

Start with AnyCap (30 Seconds)

The manual MCP path is powerful and flexible. The AnyCap path is fast and unified:

# One command, one login, every capability
npx -y skills add anycap-ai/anycap -a claude-code
anycap login

After that, ask Claude to generate an image, search the web, or store a file — and it just works. No per-server configuration. No token management across providers. No context switching.

Claude Code becomes a full multimodal agent — not just a coding assistant, but a tool that can see, create, search, and publish.