Merge pull request #360 from xJaroslav69/fix/mcp-tool-count

docs: the MCP server exposes 56 tools, not 49

README.md

@@ -177,7 +177,7 @@ omniget agent agents
 - **Multi-agent with `agent_delegate`.** One agent hands a task to another and gets the answer back as a tool result, in a child conversation you can open. A local coordinator that delegates the hard part to Claude Code costs exactly one run of the CLI.
 - **A memory the whole team shares.** `AGENTS.md` (or `CLAUDE.md`) plus markdown notes in `.omniget/kb/`, inside your project, in git if you want. The index goes into every prompt, and the agents search and write notes with `kb_search` and `kb_write`.
 - **Skills** install from a folder, a zip, or `owner/repo` and `owner/repo@skill` as in `npx skills add`, and pass through a scanner first.
-- **MCP both ways.** External MCP servers become tools, granted per agent as *auto*, *ask* or *deny*. And OmniGet serves its own 49 tools, the coding ones included, to Claude Code, Cursor, VS Code and the rest: see [OmniGet for AI agents](#omniget-for-ai-agents-mcp-server-and-claude-code-plugin).
+- **MCP both ways.** External MCP servers become tools, granted per agent as *auto*, *ask* or *deny*. And OmniGet serves its own 56 tools, the coding ones included, to Claude Code, Cursor, VS Code and the rest: see [OmniGet for AI agents](#omniget-for-ai-agents-mcp-server-and-claude-code-plugin).
 - **Context pruning, off by default.** On long conversations a judge decides which old tool outputs no longer bear on the task and replaces them with a marker; the call and its id stay, so the agent can run the tool again. It works inside a long turn as well as between turns, and on Claude Code conversations too. The local judge (MiniLM, on your machine) only ever omits an output; the optional remote judge needs a key and says plainly that previews leave your computer. Every request leaves a receipt with measured numbers: outputs omitted, estimated tokens before and after, input tokens the provider billed. No invented percentages.
 - **Budgets.** Per agent: dollars per day, tokens per turn, tool calls per turn. A spent agent goes to sit down, literally.
 
@@ -657,7 +657,7 @@ All four read the official data export LinkedIn lets you request, on your own ma
 - **Local models (Ollama).** See, download and remove local models and use them as a free provider.
 - **Humanize text.** Rewrite AI-sounding text so it reads like a person wrote it, without changing what it says. Runs on the AI provider you configured. *beta*
 - **API keys.** A local vault for keys and accounts, with a connection test, balance for OpenRouter, DeepSeek, SiliconFlow and New API, and export to Claude Code, Codex, Cherry Studio, opencode or a `.env` file.
-- **MCP server.** OmniGet's tools exposed over the Model Context Protocol on the local bridge, 49 tools behind the same token the extension uses, with ready-made config snippets for Claude Code, Claude Desktop, Cursor, VS Code, Goose and Codex. See [OmniGet for AI agents](#omniget-for-ai-agents-mcp-server-and-claude-code-plugin). *beta*
+- **MCP server.** OmniGet's tools exposed over the Model Context Protocol on the local bridge, 56 tools behind the same token the extension uses, with ready-made config snippets for Claude Code, Claude Desktop, Cursor, VS Code, Goose and Codex. See [OmniGet for AI agents](#omniget-for-ai-agents-mcp-server-and-claude-code-plugin). *beta*
 
 ### CTF and analysis (6)
 
@@ -684,7 +684,7 @@ Every tool that talks to an AI uses the provider you set in **Settings → AI**:
 
 Three doors, depending on where your agent lives.
 
-**Inside the app: the MCP server.** Turn it on in Tools → AI → MCP server and OmniGet exposes 49 of its tools over the Model Context Protocol on the same local bridge the browser extension uses, behind the same per-install token. An agent can queue a URL and watch, pause, resume or cancel it in the Downloads panel, merge, split, render, extract text from or sanitize PDFs, transcribe a file with whisper.cpp, run text to speech and OCR, resize images, find duplicates, search files, read X posts, threads, profiles, searches and trends, look up an Instagram profile through your session, compare LLM prices, humanize text, scan disks and caches, list startup items and installed apps, and pull large files or galleries with aria2 and gallery-dl. New in 0.10, the same server carries the coding tools of the built-in harness, confined to the folder you attached in the app (read, list, glob, grep, edit, write, apply a patch, run a sandboxed command, keep a plan), the project knowledge base (`kb_search`, `kb_write`) and `agent_delegate`, which hands a task to an agent of your roster. The page prints the config snippet for Claude Code, Claude Desktop, Cursor, VS Code, Goose and Codex, so it is a paste, not a setup.
+**Inside the app: the MCP server.** Turn it on in Tools → AI → MCP server and OmniGet exposes 56 of its tools over the Model Context Protocol on the same local bridge the browser extension uses, behind the same per-install token. An agent can queue a URL and watch, pause, resume or cancel it in the Downloads panel, merge, split, render, extract text from or sanitize PDFs, transcribe a file with whisper.cpp, run text to speech and OCR, resize images, find duplicates, search files, read X posts, threads, profiles, searches and trends, look up an Instagram profile through your session, compare LLM prices, humanize text, scan disks and caches, list startup items and installed apps, and pull large files or galleries with aria2 and gallery-dl. New in 0.10, the same server carries the coding tools of the built-in harness, confined to the folder you attached in the app (read, list, glob, grep, edit, write, apply a patch, run a sandboxed command, keep a plan), the project knowledge base (`kb_search`, `kb_write`) and `agent_delegate`, which hands a task to an agent of your roster. The page prints the config snippet for Claude Code, Claude Desktop, Cursor, VS Code, Goose and Codex, so it is a paste, not a setup.
 
 **The other way round: MCP servers for your agents.** In **LLM → MCP** you plug external MCP servers into OmniGet, over stdio or streamable HTTP, and their tools show up next to the built-in ones as `mcp:<server>:<tool>`. Each agent gets them one by one, as *auto*, *ask* or *deny*, and a server can never shadow a built-in tool. Secrets in a server's environment or headers are stored as references and resolved only when the server starts. Nothing connects until you enable a server.
 
@@ -832,7 +832,7 @@ Your code goes to the model you chose and nowhere else. With a local model it ne
 
 ### Can I use OmniGet as an MCP server for Claude Code, Cursor or VS Code?
 
-Yes. Tools → AI → MCP server exposes 49 tools over the Model Context Protocol, and the page prints the config snippet for Claude Code, Claude Desktop, Cursor, VS Code, Goose and Codex. The `claude-plugin/` folder also ships a Claude Code plugin that fetches and transcribes media without the desktop app. See [MCP server](#omniget-for-ai-agents-mcp-server-and-claude-code-plugin).
+Yes. Tools → AI → MCP server exposes 56 tools over the Model Context Protocol, and the page prints the config snippet for Claude Code, Claude Desktop, Cursor, VS Code, Goose and Codex. The `claude-plugin/` folder also ships a Claude Code plugin that fetches and transcribes media without the desktop app. See [MCP server](#omniget-for-ai-agents-mcp-server-and-claude-code-plugin).
 
 ### Can several AI agents work together?
 

README_pt_br.md

@@ -87,7 +87,7 @@ Novidade da 0.10 e agora o centro do app. Um comando, cerca de 15 segundos, test
 - **Claude Code, Codex e Gemini CLI com interface.** As contas entram em **LLM → Contas** com a cota na tela. Qualquer CLI que fale o Agent Client Protocol (Gemini CLI, goose, opencode) vira agente.
 - **Agentes locais com Ollama, LM Studio ou llama-server**, offline e sem chave. O `qwen3:8b` corrige o bug de demonstração em cerca de 3 minutos.
 - **Jobs, Loops e gatilhos.** O trabalho continua com a janela fechada. Um Loop repete rodadas até o seu comando de checagem (`npm test`) passar. Cron e webhook disparam jobs.
-- **Servidor MCP com 49 tools** para Claude Code, Cursor e VS Code.
+- **Servidor MCP com 56 tools** para Claude Code, Cursor e VS Code.
 - **O Mundo.** Uma casa isométrica onde cada agente anda até a sua bancada, mostra a tool num balão e acena quando precisa de você. Abra a casa, passe o código e um amigo entra para visitar.
 
 A descrição completa está no [README em inglês](README.md#agents-and-the-world).

flatpak/wtf.tonho.omniget.metainfo.xml

@@ -27,7 +27,7 @@
       <li>Instagram, TikTok, X, Reddit, Pinterest, Twitch, Bilibili and Telegram downloads</li>
       <li>156 tools in 24 categories: transcription, PDF, video, images and more</li>
       <li>AI agents with per-folder permissions, undo, jobs, loops and triggers</li>
-      <li>MCP server with 49 tools for Claude Code, Cursor and VS Code</li>
+      <li>MCP server with 56 tools for Claude Code, Cursor and VS Code</li>
       <li>The World: an isometric house where your agents are seen working</li>
     </ul>
   </description>

llms-full.txt

@@ -96,7 +96,7 @@ omniget agent agents
 - **Multi-agent with `agent_delegate`.** One agent hands a task to another and gets the answer back as a tool result, in a child conversation you can open. A local coordinator that delegates the hard part to Claude Code costs exactly one run of the CLI.
 - **A memory the whole team shares.** `AGENTS.md` (or `CLAUDE.md`) plus markdown notes in `.omniget/kb/`, inside your project, in git if you want. The index goes into every prompt, and the agents search and write notes with `kb_search` and `kb_write`.
 - **Skills** install from a folder, a zip, or `owner/repo` and `owner/repo@skill` as in `npx skills add`, and pass through a scanner first.
-- **MCP both ways.** External MCP servers become tools, granted per agent as *auto*, *ask* or *deny*. And OmniGet serves its own 49 tools, the coding ones included, to Claude Code, Cursor, VS Code and the rest: see [OmniGet for AI agents](https://github.com/tonhowtf/omniget#omniget-for-ai-agents-mcp-server-and-claude-code-plugin).
+- **MCP both ways.** External MCP servers become tools, granted per agent as *auto*, *ask* or *deny*. And OmniGet serves its own 56 tools, the coding ones included, to Claude Code, Cursor, VS Code and the rest: see [OmniGet for AI agents](https://github.com/tonhowtf/omniget#omniget-for-ai-agents-mcp-server-and-claude-code-plugin).
 - **Context pruning, off by default.** On long conversations a judge decides which old tool outputs no longer bear on the task and replaces them with a marker; the call and its id stay, so the agent can run the tool again. It works inside a long turn as well as between turns, and on Claude Code conversations too. The local judge (MiniLM, on your machine) only ever omits an output; the optional remote judge needs a key and says plainly that previews leave your computer. Every request leaves a receipt with measured numbers: outputs omitted, estimated tokens before and after, input tokens the provider billed. No invented percentages.
 - **Budgets.** Per agent: dollars per day, tokens per turn, tool calls per turn. A spent agent goes to sit down, literally.
 
@@ -510,7 +510,7 @@ All four read the official data export LinkedIn lets you request, on your own ma
 - **Local models (Ollama).** See, download and remove local models and use them as a free provider.
 - **Humanize text.** Rewrite AI-sounding text so it reads like a person wrote it, without changing what it says. Runs on the AI provider you configured. *beta*
 - **API keys.** A local vault for keys and accounts, with a connection test, balance for OpenRouter, DeepSeek, SiliconFlow and New API, and export to Claude Code, Codex, Cherry Studio, opencode or a `.env` file.
-- **MCP server.** OmniGet's tools exposed over the Model Context Protocol on the local bridge, 49 tools behind the same token the extension uses, with ready-made config snippets for Claude Code, Claude Desktop, Cursor, VS Code, Goose and Codex. See [OmniGet for AI agents](https://github.com/tonhowtf/omniget#omniget-for-ai-agents-mcp-server-and-claude-code-plugin). *beta*
+- **MCP server.** OmniGet's tools exposed over the Model Context Protocol on the local bridge, 56 tools behind the same token the extension uses, with ready-made config snippets for Claude Code, Claude Desktop, Cursor, VS Code, Goose and Codex. See [OmniGet for AI agents](https://github.com/tonhowtf/omniget#omniget-for-ai-agents-mcp-server-and-claude-code-plugin). *beta*
 
 ### CTF and analysis (6)
 
@@ -531,7 +531,7 @@ Every tool that talks to an AI uses the provider you set in **Settings → AI**:
 
 Three doors, depending on where your agent lives.
 
-**Inside the app: the MCP server.** Turn it on in Tools → AI → MCP server and OmniGet exposes 49 of its tools over the Model Context Protocol on the same local bridge the browser extension uses, behind the same per-install token. An agent can queue a URL and watch, pause, resume or cancel it in the Downloads panel, merge, split, render, extract text from or sanitize PDFs, transcribe a file with whisper.cpp, run text to speech and OCR, resize images, find duplicates, search files, read X posts, threads, profiles, searches and trends, look up an Instagram profile through your session, compare LLM prices, humanize text, scan disks and caches, list startup items and installed apps, and pull large files or galleries with aria2 and gallery-dl. New in 0.10, the same server carries the coding tools of the built-in harness, confined to the folder you attached in the app (read, list, glob, grep, edit, write, apply a patch, run a sandboxed command, keep a plan), the project knowledge base (`kb_search`, `kb_write`) and `agent_delegate`, which hands a task to an agent of your roster. The page prints the config snippet for Claude Code, Claude Desktop, Cursor, VS Code, Goose and Codex, so it is a paste, not a setup.
+**Inside the app: the MCP server.** Turn it on in Tools → AI → MCP server and OmniGet exposes 56 of its tools over the Model Context Protocol on the same local bridge the browser extension uses, behind the same per-install token. An agent can queue a URL and watch, pause, resume or cancel it in the Downloads panel, merge, split, render, extract text from or sanitize PDFs, transcribe a file with whisper.cpp, run text to speech and OCR, resize images, find duplicates, search files, read X posts, threads, profiles, searches and trends, look up an Instagram profile through your session, compare LLM prices, humanize text, scan disks and caches, list startup items and installed apps, and pull large files or galleries with aria2 and gallery-dl. New in 0.10, the same server carries the coding tools of the built-in harness, confined to the folder you attached in the app (read, list, glob, grep, edit, write, apply a patch, run a sandboxed command, keep a plan), the project knowledge base (`kb_search`, `kb_write`) and `agent_delegate`, which hands a task to an agent of your roster. The page prints the config snippet for Claude Code, Claude Desktop, Cursor, VS Code, Goose and Codex, so it is a paste, not a setup.
 
 **The other way round: MCP servers for your agents.** In **LLM → MCP** you plug external MCP servers into OmniGet, over stdio or streamable HTTP, and their tools show up next to the built-in ones as `mcp:<server>:<tool>`. Each agent gets them one by one, as *auto*, *ask* or *deny*, and a server can never shadow a built-in tool. Secrets in a server's environment or headers are stored as references and resolved only when the server starts. Nothing connects until you enable a server.
 
@@ -663,7 +663,7 @@ Your code goes to the model you chose and nowhere else. With a local model it ne
 
 ### Can I use OmniGet as an MCP server for Claude Code, Cursor or VS Code?
 
-Yes. Tools → AI → MCP server exposes 49 tools over the Model Context Protocol, and the page prints the config snippet for Claude Code, Claude Desktop, Cursor, VS Code, Goose and Codex. The `claude-plugin/` folder also ships a Claude Code plugin that fetches and transcribes media without the desktop app. See [MCP server](https://github.com/tonhowtf/omniget#omniget-for-ai-agents-mcp-server-and-claude-code-plugin).
+Yes. Tools → AI → MCP server exposes 56 tools over the Model Context Protocol, and the page prints the config snippet for Claude Code, Claude Desktop, Cursor, VS Code, Goose and Codex. The `claude-plugin/` folder also ships a Claude Code plugin that fetches and transcribes media without the desktop app. See [MCP server](https://github.com/tonhowtf/omniget#omniget-for-ai-agents-mcp-server-and-claude-code-plugin).
 
 ### Can several AI agents work together?
 

llms.txt

@@ -1,6 +1,6 @@
 # OmniGet
 
-> OmniGet is a free, open source (GPL-3.0) desktop app for Windows, macOS and Linux with two halves. It downloads online courses (Udemy, Hotmart, Kiwify, Rocketseat), YouTube, Instagram, TikTok, X/Twitter and about 1,800 other sites through a graphical interface over yt-dlp, then plays, reads, transcribes and converts what was saved. And it runs AI coding agents on the user's own machine: Claude Code, Codex, Gemini CLI and local Ollama models, with per-folder permissions, undo, durable jobs, loops that repeat until a test command passes, an embedded MCP server with 49 tools, and an isometric house called the World where the agents are seen working.
+> OmniGet is a free, open source (GPL-3.0) desktop app for Windows, macOS and Linux with two halves. It downloads online courses (Udemy, Hotmart, Kiwify, Rocketseat), YouTube, Instagram, TikTok, X/Twitter and about 1,800 other sites through a graphical interface over yt-dlp, then plays, reads, transcribes and converts what was saved. And it runs AI coding agents on the user's own machine: Claude Code, Codex, Gemini CLI and local Ollama models, with per-folder permissions, undo, durable jobs, loops that repeat until a test command passes, an embedded MCP server with 56 tools, and an isometric house called the World where the agents are seen working.
 
 OmniGet is for two kinds of people. People who want a downloader: a Udemy downloader, a Hotmart downloader, a YouTube downloader, a yt-dlp GUI, an Instagram, TikTok, X, Reddit, Pinterest, Twitch, Bilibili or Telegram downloader, a torrent client, with no terminal. And people who want a desktop app for AI agents: a GUI for Claude Code and Codex, a client for any Agent Client Protocol (ACP) agent, local agents on Ollama or LM Studio, background jobs, scheduled runs, and an MCP server for Claude Code, Claude Desktop, Cursor, VS Code, Goose and Codex.
 
@@ -15,7 +15,7 @@ OmniGet is for two kinds of people. People who want a downloader: a Udemy downlo
 - Cost: free, no account, no ads, no paid tier, no telemetry
 - Built with: Tauri 2, Rust, SvelteKit (Svelte 5), SQLite
 - Engines it manages for the user: yt-dlp (verified by SHA-256 before it runs), FFmpeg, aria2, gallery-dl, whisper.cpp
-- Size: 156 tools in 24 categories, 49 MCP tools, 4 official plugins, 14 themes, 12 interface languages
+- Size: 156 tools in 24 categories, 56 MCP tools, 4 official plugins, 14 themes, 12 interface languages
 - Community: https://discord.gg/jgdxyPy7Vn
 
 ## Downloading
@@ -39,10 +39,10 @@ OmniGet is for two kinds of people. People who want a downloader: a Udemy downlo
 - The World: an isometric house, rendered in WebGL2 from a Rust simulation, where each agent walks to its own workbench, shows the running tool in a balloon and waves when it needs permission. Inspired by generative agents and ai-town, driven by real work instead of a script. 64 fps median with eight agents working (debug build).
 - Visits: "Open the house" gives a code; a friend's app connects through a relay (`wss://chat.tonho.wtf/v1/room`, or a self-hosted `omniworld-server`). The host app is the authority and keys never leave the machine.
 
-## MCP server (49 tools)
+## MCP server (56 tools)
 
 - [MCP server and Claude Code plugin](https://github.com/tonhowtf/omniget#omniget-for-ai-agents-mcp-server-and-claude-code-plugin): How to connect.
-- Turn it on in Tools → AI → MCP server. It serves 49 tools over the Model Context Protocol on the local bridge (127.0.0.1), behind a per-install token. The page prints ready-made config for Claude Code, Claude Desktop, Cursor, VS Code, Goose and Codex.
+- Turn it on in Tools → AI → MCP server. It serves 56 tools over the Model Context Protocol on the local bridge (127.0.0.1), behind a per-install token. The page prints ready-made config for Claude Code, Claude Desktop, Cursor, VS Code, Goose and Codex.
 - Tools include: queue and control downloads, PDF merge/split/render/extract/sanitize, whisper.cpp transcription, text to speech, OCR, image resize, duplicate finder, file search, X posts/threads/profiles/search, Instagram profile lookup, LLM price comparison, disk and cache scans, aria2 and gallery-dl fetches, the coding tools of the harness, `kb_search`, `kb_write` and `agent_delegate`.
 - OmniGet is also an MCP client: external MCP servers (stdio or streamable HTTP) become agent tools, granted per agent as auto, ask or deny.
 - A Claude Code plugin in `claude-plugin/` adds `/omniget:fetch`, `/omniget:transcribe` and `/omniget:research` without the desktop app.
@@ -76,7 +76,7 @@ OmniGet is for two kinds of people. People who want a downloader: a Udemy downlo
 - Can an AI agent loop until the tests pass? Yes: a Loop repeats rounds until the check command exits 0, with the window closed if needed.
 - Does it work with Ollama only, offline? Yes: a fresh install is local first and needs no key.
 - Is my code sent anywhere? Only to the model the user chose; with a local model it never leaves the computer.
-- Can Claude Code, Cursor or VS Code use OmniGet's tools? Yes, through the built-in MCP server with 49 tools.
+- Can Claude Code, Cursor or VS Code use OmniGet's tools? Yes, through the built-in MCP server with 56 tools.
 - Does it resume interrupted downloads? Yes, with retry and backoff on rate limits.
 
 ## Privacy and scope