Repolode
Zen4-bit

Zen4-bit/Proxima

A local MCP server that routes prompts from Cursor, VS Code, Windsurf or Claude Desktop to ChatGPT, Claude, Gemini, Perplexity, BYOK APIs or local Ollama models.

Stars
1,198
Forks
163
Open issues
28
Last commit
3 days ago

What is it?

Proxima is a local gateway that exposes several AI models as one MCP server and OpenAI-compatible HTTP endpoint on 127.0.0.1. Coding clients such as Cursor, VS Code, Windsurf and Claude Desktop connect to it over stdio and can call tools like ask_claude, ask_gemini, smart_query or deep_search.

There are two routing modes. Session mode sends prompts through the user's existing logged-in browser sessions for ChatGPT, Claude, Gemini and Perplexity, without storing passwords. BYOK mode uses official API keys for providers such as OpenAI, Anthropic, Google, DeepSeek, Groq and Mistral, and can also connect to local models served by Ollama or LM Studio. BYOK keys are stored through the operating system's keychain via Electron SafeStorage.

The project also includes proxima-agent, a Python assistant that edits code, runs tests and drives Chrome over CDP, with permission modes from full auto to suggest-only. It keeps SQLite databases for conversation history, learned fixes and reusable skills. A desktop app, CLI, REST API, WebSocket interface and Python and JavaScript SDKs are provided.

Key features

  • 01Exposes ChatGPT, Claude, Gemini and Perplexity to editors as MCP tools via browser sessions.
  • 02Supports BYOK routing to OpenAI, Anthropic, Google, DeepSeek, Groq, Mistral and other providers.
  • 03Connects to local models running on Ollama or LM Studio.
  • 04Offers multi-model tools such as smart_query, verify and crew.
  • 05Includes an OpenAI-compatible REST API and WebSocket streaming on localhost.
  • 06Provides a Python agent that runs tests and edits code with gated approval modes.

Use cases

  • Letting Cursor or VS Code agents query several frontier models without buying separate API plans.
  • Routing coding questions to a fully local Ollama model so code stays on the machine.
  • Cross-checking a model's answer against another provider with the verify or consensus tools.
  • Adding one local OpenAI-compatible endpoint that existing tools can point to.

Things to know

  • The README lists a Non-Commercial license, and the repository metadata shows NOASSERTION, so commercial use needs a license check.
  • Session Mode automates logged-in browser accounts, which may fall outside some providers' terms of service for programmatic access.
  • Full use of the BYOK mode requires your own API keys for the providers you want to use.

Topics

FAQ

What is Proxima?
A local MCP server that routes prompts from Cursor, VS Code, Windsurf or Claude Desktop to ChatGPT, Claude, Gemini, Perplexity, BYOK APIs or local Ollama models. Proxima is a local gateway that exposes several AI models as one MCP server and OpenAI-compatible HTTP endpoint on 127.0.0.1. Coding clients such as Cursor, VS Code, Windsurf and Claude Desktop connect to it over stdio and can call tools like ask_claude, ask_gemini, smart_query or deep_search.
Is Proxima free to use? What license does it have?
Proxima is open source under the Other license. Check the license terms before using it in commercial products.
Is Proxima still maintained?
Actively maintained. The last commit to Proxima was on Oct 16, 2026.
What can I use Proxima for?
Letting Cursor or VS Code agents query several frontier models without buying separate API plans. Routing coding questions to a fully local Ollama model so code stays on the machine. Cross-checking a model's answer against another provider with the verify or consensus tools. Adding one local OpenAI-compatible endpoint that existing tools can point to.
Where can I get Proxima?
The source code and installation instructions are on GitHub: https://github.com/Zen4-bit/Proxima

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects