glossary
Plain-English definitions of every term in the Frenchie docs and marketing pages. Useful if MCP, sync vs async, or skill packs are new to you — and useful for your agent if it's asked "what does X mean?"
Model Context Protocol — the open standard for how AI agents discover and call external tools. Frenchie ships as an MCP server so any MCP-compatible agent (Claude Code, Cursor, Codex, Antigravity, Claude Desktop, and others) can call OCR, transcription, and image generation through one consistent interface.
A program that exposes a set of tools and resources to an AI agent over the Model Context Protocol. Frenchie runs in two modes: a local stdio MCP server installed via npx for desktop agents, and an HTTP MCP endpoint at mcp.getfrenchie.dev for hosted/web agents.
MCP transport where the agent spawns the server as a local subprocess and communicates over stdin/stdout JSON-RPC. Frenchie's primary integration path. Auto-saves results to .frenchie/<name>/ next to the source file because the server runs on the same machine.
MCP transport where the agent calls a remote HTTP endpoint. Frenchie's hosted/web fallback at mcp.getfrenchie.dev — used when the agent can't spawn local binaries (Lovable, Manus, Claude.ai, ChatGPT.com). Requires the upload → reference flow because the server has no access to local files.
Optical Character Recognition — converting PDFs and images into machine-readable text. Frenchie returns clean Markdown with table structure preserved and figures pulled out as separate PNG files. 1 credit per page. Supported formats: PDF, PNG, JPG, WebP.
Converting audio or video into text. Frenchie returns Markdown transcripts with paragraph structure. 2 credits per minute. Supported formats: MP3, M4A, WAV, MP4, MOV, WebM. 50+ languages including Thai, Japanese, Chinese, Arabic, French, Spanish.
Producing an image file from a text prompt. Frenchie generates one image per call via gpt-image-2. 20 credits per image. Output formats: PNG (default), JPEG, WebP. Sizes: 1024x1024, 1536x1024, 1024x1536, or auto. Rate limit: 50 images per hour, 250 per day per user.
A free installable bundle that teaches an agent how to use Frenchie's MCP tools. Ships slash commands like /ocr, /transcribe, /generate-image, and /frenchie-status. Distributed via the @lab94/frenchie npm package — install with `npx @lab94/frenchie install`.
Frenchie's routing for small jobs that complete inline — typically OCR on short PDFs and short audio transcription. The MCP tool call returns the result in one response, no polling. Anything large enough to take more than a few seconds is routed to the async path automatically.
Frenchie's routing for jobs too large to complete inline — long audio, big PDFs. The MCP tool returns a job ID immediately; the agent polls with `get_job_result` (or waits for a smart-wait window of up to 90 seconds). Lets your agent keep working on other things while the job runs.
Frenchie's billing unit. $1 buys 100 credits. 1 credit per OCR page, 2 credits per transcription minute, 20 credits per generated image. 100 free credits on signup, no card required. Credits don't expire and refund automatically when a job fails.
Frenchie deletes processed files immediately and expires the result payload 30 minutes after first delivery. By design — Frenchie is a processing utility, not a storage product. Agents that need to revisit a result should save the Markdown locally before the window closes.
Every Frenchie capability returns the same envelope: `{ status, jobId, creditsUsed, resultExpiresAt, result: { kind: 'markdown' | 'image', ... } }`. Agents branch on `result.kind` instead of guessing per-tool output shape. Introduced in v0.4.0 alongside image generation.
An agent with first-class auto-install support: the `npx @lab94/frenchie install` command auto-detects the agent and writes the right MCP config. Currently: Claude Code, Cursor, Codex, Antigravity, and Claude Desktop. Other MCP clients work too via `--agent <name>`.
