Oracle is a command-line tool and Model Context Protocol (MCP) server that bundles a prompt together with the files you select, sends that context to an AI model through a provider API or a signed-in browser session, and stores the result as a session. It is aimed at developers and coding agents that want a second-model review grounded in the actual project.
Three paths are available. API mode calls provider endpoints and supports OpenAI, Azure OpenAI, Anthropic, Gemini, xAI, OpenRouter, and compatible endpoints. Browser mode drives a signed-in ChatGPT session through Chrome automation, or Gemini through a cookie-based client, so no API key is needed. Render mode assembles the bundle and prints it for you to inspect, copy, or paste yourself.
Context is composed with repeated --file arguments that accept files, directories, globs, and ! exclusions. A dry run reports the resolved file list and a token estimate before anything is sent, and generated text bundles carry stable line numbers so an answer can cite path:line. In browser mode a single text file stays native, while multiple text or source files are packed into one flattened bundle or a ZIP.
Runs are stored locally, so a long response can finish in the background and a finished answer can be replayed, reattached to, restarted, or continued with more context through a follow-up. The --models flag runs a panel of several models in one command and records per-model usage, cost, output, and partial failures in the same session.
For agent use, an oracle-mcp stdio server and an installable skill wire it into Claude Code, Codex, Cursor, and other MCP clients, with plain stdout JSON envelopes for scripting.
Features
- Prompt and file bundles: a prompt is packed with selected files, controlled by globs, exclusions, and size guards
- Three engines: provider APIs, a signed-in browser session over Chrome, or a rendered bundle you paste yourself
- Provider coverage: OpenAI, Azure OpenAI, Anthropic, Gemini, xAI, OpenRouter, and compatible endpoints
- Multi-model panels: --models runs several models in one command and records per-model usage, cost, output, and partial failures
- Sessions and follow-ups: runs persist on disk for replay, reattachment, restart, or continuation from a stored session or response id
- Citable output: stable line numbers in generated bundles let an answer reference path:line
- Dry runs: a file report and token estimate show the resolved context before a request is sent
- Readiness checks: oracle doctor --providers inspects provider and key readiness without exposing credentials
- Agent integration: an oracle-mcp stdio server plus a skill for Claude Code, Codex, Cursor, and other MCP clients
