AI Assistant
An in-app AI assistant and an MCP server that let a model read and analyze your captured traffic.
Traffic Jam ships a shared tool layer — a set of read, replay, analysis, and authoring actions over the backend HTTP API and the analysis libraries. Two consumers use the exact same tools:
- The in-app assistant (/docs/en/ai-assistant/assistant/): a chat panel in the API lab that talks to any OpenAI-compatible model endpoint you configure.
- The MCP server (/docs/en/ai-assistant/mcp/): a Model Context Protocol endpoint so external clients — Claude Code, Claude Desktop, or any MCP host — can drive Traffic Jam from outside the app.
Both expose the same capabilities: list and inspect captures, decode bodies, replay requests, minimize and mutate-probe them, run the collector analyses (conformance diff, signature cracking, token lifecycle), and persist parser artifacts — decode scripts, decode pipelines, and session recipes — so a model can close the loop from “I figured out the codec” to “here is the stored parser.”
[!NOTE] The assistant and MCP tools read, replay, analyze, and author parser artifacts. They cannot start captures, upload client certificates, or delete data. Replay tools send live traffic to the real endpoints they were captured from — the in-app assistant asks for confirmation before each mutating tool by default. Authoring tools (
save_decode_script,save_decode_pipeline,save_recipe) only write to the local project database and send no traffic.