Some tools require MCP. Most don't. Knowing the difference keeps your agent sharp.
tokens consumed by 4 MCP servers before typing a single prompt
tool definitions loaded, most of which the agent never called
context threshold where Claude Code auto-defers MCP tool definitions
When MCP provides a capability your agent cannot get any other way.
Your agent was trained on millions of examples for these tools. Zero token overhead.
Skills are not a replacement for MCP. They add metadata, guardrails, and project-specific context to any tool.
Describe which of 40+ tools matter. Document gotchas, recommended sequences, project-specific context.
Add project-specific flags, guardrails, API specs. Point to OpenAPI docs, explain which endpoints to avoid.
Wrap CLIs with safety checks, rate limits, and defaults. More work than MCP, but real guardrails.
For every MCP server in your config, run it through this.
Use project-scoped .mcp.json so each server only loads where needed.
Tool usage follows a power law. 2-3 tools do the work. The rest eat context.
Treat .mcp.json like package.json. Every server addition is a conversation.
MCP servers are local subprocesses with full filesystem access. No sandboxing by default.
The question is whether each server is earning its context cost.