Every server in the config talks on every call.
The lesson 38.4 chatbot reads server_config.json, connects to each server, and sends every tool it finds to the model on every request. Switch servers on and off. The tool lists are the real list_tools replies, recorded with MCP SDK 1.30.0.
- Tool definitions per call
- Characters of tool JSON
- About this many tokens
- Names sent twice
What each server sends and receives
One row per server
| Server | Tools sent | Characters | Calls it gets | Tool names |
|---|---|---|---|---|
| Every request |
Characters count each definition the chatbot builds (name, description, input_schema) as compact JSON. Tokens use the four-characters-per-token rule of thumb. “Calls it gets” counts the tool names that tool_to_session routes to that server.
One call through the router
Results recorded from @modelcontextprotocol/server-filesystem 2.1.0. research_server.py sits in the project folder, which the page calls mcp_project; handbook/ is a folder inside it.