Tool Interfaces and MCP¶
A tool's real cost is measured in tokens, not lines of code — every schema sits in context on every turn, whether or not it's ever called.
flowchart LR
J["Junior: what MCP is, one round trip"] --> M["Middle: MCP vs CLI on token cost"]
M --> S["Senior: choose the boundary, add guardrails"]
S --> P["Professional: govern the tool registry"]
Levels¶
| Level | Guide | You are done when |
|---|---|---|
| Junior | What MCP is, one round trip | You can explain MCP's client-server model and trace one tool call from discovery to result. |
| Middle | MCP vs CLI on token cost | You can compare a CLI tool and an MCP server on token cost, control, and guardrails for a concrete case. |
| Senior | Choose the boundary, add guardrails | You can decide which capabilities deserve a typed MCP server vs. a generic shell tool, with cost caps and failure semantics. |
| Professional | Govern the tool registry | You can run a shared tool registry with schema versioning and security review at 100+ tools. |
Practice rule¶
Before adding an MCP server or a new tool schema, ask what it costs on every turn it's not used, and what capability it adds that the model's existing tools (a shell, a CLI it already knows) don't already provide.
Related¶
- Search and Retrieval — search itself is usually exposed to the model as one of these tool interfaces.
- Context Fundamentals — the budget that tool schemas compete against, whether called or not.
- Agent Workflow — the loop that decides when to call a tool at all.