MCP Overview
AI & MCP: give your AI agent network control, HTTPeep's built-in MCP server lets AI agents read live traffic, modify forward rules, manage DNS, and control the proxy — not just observe it.
HTTPeep includes a built-in MCP (Model Context Protocol) server that exposes the full proxy engine to AI agents. Unlike read-only packet capture tools, HTTPeep MCP lets agents take action: start and stop the proxy, rewrite forward rules, flip DNS override environments, manage bypass lists, and more — all without you touching the GUI or terminal yourself.
What agents can do
When you connect an agent to HTTPeep via MCP, it gains access to nine capability groups:
- Query live traffic — list sessions, drill into request/response detail, tail logs, and poll events in real time
- Generate and modify rules — add, update, or replace forward rules based on observed traffic patterns
- Control DNS overrides — switch environment groups, add host entries, or replace the entire DNS config
- Manage bypass and downstream lists — set bypass mode, test domains, configure downstream routing
- Start, stop, and restart the proxy — including pause/resume for noise reduction during focused operations
- Inspect and install certificates — check root CA trust status and trigger installation
- Configure system proxy settings — enable or disable system-wide proxy interception per protocol
- Modify MCP settings — update the MCP config, start or stop the LAN MCP service, repair CLI PATH
The AI debugging workflow
The most powerful use of HTTPeep MCP is a continuous loop: the agent sees real traffic, reasons about it, and makes changes — all in one conversation.
The agent calls httpeep.sessions_list with a filter such as status:>=500 to find failing requests in the last hour.
Using httpeep.session_detail, it inspects request headers, response bodies, and timing to identify the root cause.
It adds a forward rule via httpeep.rules_*, adjusts a DNS override via httpeep.dns_*, or updates bypass behavior via httpeep.bypass_*.
It calls httpeep.proxy_restart to apply the changes, then re-queries sessions to confirm the fix.
This replaces a workflow that previously required switching between the GUI, terminal, and your editor.
Transport modes
HTTPeep MCP supports two transport modes. Choose based on where your agent runs.
The agent spawns httpeep-cli mcp serve as a subprocess and communicates over stdin/stdout.
When to use: Any agent running on the same machine as HTTPeep — Codex, Claude Code, Kimi, OpenCode, and most desktop-based agents.
Why it's preferred:
- No port to configure or expose
- Not affected by firewalls or port conflicts
- Simpler configuration — one command, no addresses
httpeep-cli mcp serveYour agent configuration handles launching this command automatically. You don't run it manually.
Prerequisites
Before connecting an agent, verify that everything is in order:
The MCP server is part of the HTTPeep proxy engine. The desktop app must be running for any MCP tool calls to work.
httpeep-cli --versionIf the command is not found, open the desktop app, go to Settings → MCP, and click Repair CLI / PATH. Then restart your terminal.
httpeep-cli mcp doctorThis checks that the CLI is reachable, the proxy engine is running, and the MCP service is responding correctly.
Run httpeep-cli mcp doctor any time a tool call fails unexpectedly. It will tell you exactly which component is unreachable.
Ready to connect your agent? See Setup for per-agent configuration instructions.
HTTPeep CLI Skill for AI Agents
Install the HTTPeep CLI Skill so AI coding agents can inspect real HTTP traffic, debug API failures, and close the loop between code changes and network evidence.
MCP Setup
Set up AI agent integration with HTTPeep MCP, Step-by-step setup for Codex, Claude Code, Kimi, and OpenCode — including config files, CLI commands, and how to fix common PATH issues.