AI Setup

gori has three entry points over one project and one engine: gori (the TUI, for you), gori run (the headless CLI, for scripts), and gori mcp (the MCP server, for AI agents). This page is the AI path. It gets an agent connected to a gori project and running its first request.

There is no chat window inside the TUI. You bring your own model and client; gori exposes the project over a clean tool interface, and the agent reads your traffic and drives the same tools you do. For the full tool catalog and deeper topics (live intercept, the design rationale), see the MCP Server guide.

Before you begin. Install gori and have an MCP-capable client ready (Claude Code, Claude Desktop, OpenAI Codex, Antigravity, Grok, and others).

1. Install the server into your client

gori mcp speaks JSON-RPC 2.0 over stdio: an AI client spawns it, sends requests on STDIN, and reads results on STDOUT (STDERR carries logs). Rather than hand-editing each client's config, let gori write it for you:

gori mcp --install-claude-code   # Claude Code
gori mcp --install-claude        # Claude Desktop
gori mcp --install-codex         # OpenAI Codex
gori mcp --install-agy           # Antigravity CLI
gori mcp --install-grok          # Grok
Flag Client Config written
--install-claude Claude Desktop ~/Library/Application Support/Claude/claude_desktop_config.json (macOS)
--install-claude-code Claude Code ~/.claude.json (mcpServers.gori)
--install-codex OpenAI Codex ~/.codex/config.toml ([mcp_servers.gori])
--install-agy Antigravity CLI ~/.gemini/antigravity-cli/mcp_config.json
--install-grok Grok ~/.grok/config.toml ([mcp_servers.gori])

Each command prints the file it wrote and the exact launch command it recorded. Codex and Grok use a TOML [mcp_servers.gori] table, not JSON. Restart the client (or reopen the session) afterward so it reloads its MCP servers.

Wiring it up by hand instead? The server is just the gori mcp command over stdio. Point any MCP client at that command with no extra arguments.

Checkpoint. Your client lists gori's tools (list_history, send_request, project_info, and the rest). If they do not appear, confirm the client was restarted and that gori is on the PATH the client uses.

2. Pin the right project

Each gori project is its own database, so the agent has to serve the right one. With no selector, gori path-binds the nearest Git workspace to its own project; started outside a Git workspace it fails closed rather than serving an unrelated project. If your client launches MCP servers from a fixed directory, pin the project when you install:

gori mcp --project my-engagement --install-codex     # a named project's database
gori mcp --db /path/to/project.db --install-claude-code   # a specific database file

Then have the agent call project_info first. It reports the selected project, the database path, the workspace root, and how the selection was made, so the agent confirms it is looking at the right data before touching anything. The full selection rules live in Choosing a Project.

3. Hand off safely with read-only

By default the server also exposes action tools that send live requests and write issues. To give an agent (or a teammate you do not fully trust with the target) only the read tools, install it read-only:

gori mcp --read-only --install-claude-code

Read-only keeps list_history, get_flow, list_sitemap, and the other inspection tools while disabling send_request, issue writes, and the intercept mutators. Pure-compute helpers like decode and jwt_decode stay available because they never touch the network or your data.

4. Drive your first request

With the tools live, prompt the agent in plain language. It maps your intent onto the read and action tools:

"List the last 20 POSTs to /login, resend the newest one with a different password, and open an issue if the status code changes."

A capable agent turns that into a short tool sequence:

β†’ list_history   method:POST path~/login   (newest 20)
β†’ get_flow       <the newest flow>
β†’ send_request   POST /login  (edited body)
β†’ create_issue   "Auth bypass on /login" severity:high

Agent actions are not silent. Each one lands in gori's notification center tagged as agent-driven and rendered differently from your own actions, so you can see what a co-pilot did to a project while you were reading another tab.

Next Steps