Part of Extend Claude With Knowledge Work Skills
Engineering Blueprint
Connect to External MCP Servers
Universal client for integrating MCP servers (Zapier, GitHub, Sequential Thinking) without context bloat.
What problem does this solve?
Gives Claude on-demand access to external MCP servers by loading tool schemas progressively instead of dumping thousands of tokens of tool definitions into context upfront. You configure servers once (with API keys), then load only the tools you need when you need them through three simple commands: list servers, list tools from a server, or call a specific tool.
How does it work?
Set up once by creating or editing .claude/skills/mcp-client/references/mcp-config.json with your MCP servers (Zapier, GitHub, Sequential Thinking, etc.) and API keys. Then run commands as needed: 1. python .claude/skills/mcp-client/scripts/mcp_client.py servers to list configured servers. 2. python .claude/skills/mcp-client/scripts/mcp_client.py tools <server> to list available tools and their schemas. 3. python .claude/skills/mcp-client/scripts/mcp_client.py call <server> <tool> '{json}' to execute a tool. Output: Progressively disclosed tool schemas, avoiding context window bloat while providing full integration capability.
Included Skills
No skills in this group
What's required to run this?
Requires Python 3.11+ and MCP SDK (pip install mcp fastmcp). Config format matches Claude Desktop's MCP config. API keys stored in .mcp-config.json must be treated as secrets. Transport type is auto-detected from URL format (API key for Zapier, command+args for local stdio servers, /sse or /mcp suffixes for alternate transports).
Key Features
Progressive Tool Loading
Load tool schemas only when you request them, not all at once. Keeps context efficient.
Multiple Transport Types
Supports remote servers via API key (Zapier), stdio servers (local tools), SSE, and HTTP transports.
Tool Gotchas Documentation
Tests tools once and documents quirks in CLAUDE.md, saving context and preventing repeated errors.
Flexible Configuration
Reads config from environment variable, skill config file, project root, or home directory in that order.
Python-Based Execution
Scripts can run without loading into context, enabling deterministic execution and parameter injection.
Tools in this Blueprint
About This Blueprint
- Industry
- Technology
More Blueprints to explore
Answer pre-purchase questions to convert hesitant shoppers with AI
Shoppers get accurate answers about fit, compatibility, ingredients, and delivery even when your team is offline, so fewer leave before checkout. Conversation insights also reveal which product-page details could prevent the next unanswered question.
Urska B.
GTM Lead
Find the root cause behind repeat support tickets with AI
Catch a few unusual tickets before a product or fulfillment issue turns into hundreds of refunds, chargebacks, and bad reviews. Cluster transcript evidence by SKU, order date, location, and carrier so operations can trace the cause and support can reach affected customers early.
Urska B.
GTM Lead