Turn Free Web AIs (DeepSeek, Gemini, ChatGPT, Claude) into a Local MCP Agent Harness
A raw LLM in a browser tab is just a text predictor—what turns models like DeepSeek, Gemini, ChatGPT, or Claude into real autonomous agents is the agent harness around them. I built Web MCP Bridge to wrap their free web interfaces in a complete local Model Context Protocol (MCP) harness using a Manifest V3 extension and a lightweight Python WebSocket bridge ( 127.0.0.1 ). It lets the web AI read…
DeepSeek, Gemini, ChatGPT, and Claude are large language models (LLMs) that can be transformed into autonomous agents through an agent harness. The web interface of these models can be wrapped in a Model Context Protocol (MCP) harness using a Manifest V3 extension and a lightweight Python WebSocket bridge. This enables the web AI to read files, run terminal commands, and interact with local stdio MCP servers without requiring any API keys, subscriptions, or local GPUs.
In the most recent version, version 1.8.0, the bridge has been optimized to handle hundreds of tools without freezing browser tabs. This is achieved by replacing massive 100k-character schema dumps with a more compact ~2k-character paginated catalog that the models can query on demand. Additionally, a 3-layer self-healing JSON parser has been introduced to fix broken model outputs such as unescaped Windows paths or trailing commas.
The bridge also features a redesigned bilingual UI, supporting both English and Spanish, and dual licensing under MIT and Apache 2.0. All operations are performed locally on the user's machine, with the bridge rejecting non-extension origins, isolating tool outputs to prevent prompt injection, and defaulting to "Ask every time" to ensure no action runs without the user's explicit approval. The bridge's code and quick-start guide are available on GitHub: AstraxOf/web-mcp-bridge.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.