>
Run ChatGPT,
Claude,
Gemini &
Perplexity inside Cursor, VS Code, Windsurf & Antigravity
Free Local Multi-AI MCP Server & Autonomous Action Agent — Zero API Key Costs
Independent community rankings and trending milestones tracked on GitHub and Trendshift.
Featured on GitHub Trending, first reaching #3 on May 7, 2026.
Ranked on Trendshift's daily list for JavaScript, first achieved on 12th, Apr 2026.
Proxima is open-source and supported by the developers and sponsors who help fund active development.
Choose between 100% free browser sessions, your own custom API keys, or completely offline local hardware.
Automates your active browser logins inside sandboxed background views. Zero API keys, zero token fees, unlimited coding.
Direct official API access for production pipelines. Encrypted securely in your operating system's native keychain vault.
Run open-weight models on your local GPU or CPU. Your code never leaves your computer — zero internet required.
Route prompts to the best model for the task. Zero friction.
AI coding agents hallucinate, guess outdated APIs, and struggle with newer frameworks — because they're stuck with one model's old training data. Proxima lets multiple AI models with web access and stronger reasoning work together, cross-verify each other's answers, and produce results no single model can match alone.
OpenAI-compatible endpoint at localhost:3210 (the port is auto-discovered if it is busy). Stream responses in real-time via WebSocket, set thinking depth, and cancel in-flight requests.
list_available_models, deep_search, review_code, security_audit, write_tests, fix_error, and 35 more — ready to use in Cursor, VS Code, Windsurf, Antigravity, or Claude Desktop.
ask_all_ais, smart_query, crew, run_workflow, debate — send one prompt across multiple AI providers at once, cross-verify answers, or chain outputs between role-based agents.
Run "proxima ask", "proxima fix", "proxima code" from any terminal. Supports pipes — send build errors or git diffs straight to AI.
Everything on localhost. No cloud relay. Works headless in background — MCP tools keep running even without the app window open.
Turns prompts into real-world action. Executes terminal tasks, creates scripts, manages files, and completes multi-step goals across your computer.
Automate virtually any task across your entire system — from browser actions and web research to custom scripts, API data pipelines, and background workflows.
An autonomous local action engine that completes tasks, automates workflows, and handles complex computer operations hands-free — runs without API keys, with custom API keys, or completely offline with local models.
Takes your high-level goals and executes them end-to-end — running commands, managing files, processing data, and writing code.
Navigates the web, extracts live research, interacts with online services, and automates browser actions directly through Chrome.
Automatically catches terminal errors, analyzes what went wrong, and fixes issues iteratively until the task is successfully finished.
Remembers your workspace preferences, learns from past workflows, and creates reusable automation shortcuts over time.
Pick models from what your accounts actually offer, control reasoning depth, cancel runaway requests, and stop worrying about busy ports.
The new list_available_models tool and
proxima://models resource report the models your logged-in ChatGPT and Gemini
sessions and BYOK providers can serve right now. Ask for one you do not have and you get a
clear error, not a silent swap.
Set thinking_level to low, medium or high from MCP,
REST, the SDKs or --thinking on the CLI, and stop an in-flight request with
POST /v1/cancel.
If port 3210 is taken, the gateway moves to the next free port and
records it in rest-port.json. The CLI, SDKs and agent find it on their own. A
new ES module build of the JavaScript SDK ships alongside CommonJS.
Deep technical architectural guides, Model Context Protocol internals, and autonomous agent loops.
Learn how JSON-RPC 2.0 schemas, stdio streams, and dynamic tool dispatchers route 41 tools across Cursor, VS Code, and Windsurf.
How sandboxed Electron partitions, WebSocket token stream capture, and SafeStorage key encryption enable local multi-model AI coding.
Discover how autonomous coding agents capture stack traces, inspect AST symbol maps, and iteratively patch code using SQLite memory.
Experience seamless MCP tool calling, OpenAI-compatible REST endpoints, and terminal CLI workflows in real-time.
Switch between providers instantly. Use the best model for each task.
Best for code generation, debugging, and structured output
Best for deep reasoning, long code reviews, and analysis
Best for multimodal tasks, image understanding, and research
Best for real-time research, current docs, and citations
Everything you need for AI-powered development. The list below is the exact set the MCP server registers in v5.5.0, plus the proxima://status and proxima://models resources.
list_available_models
Discover live models per account
ask_chatgpt
Query ChatGPT
ask_claude
Query Claude
ask_gemini
Query Gemini, with thinking depth
ask_perplexity
Query Perplexity (web + citations)
ask_model
Any session or BYOK provider
ask_all_ais
Same prompt to every provider
smart_query
Auto-route: verify, consensus, collaborate
new_conversation
Reset a provider's context
generate_code
Write new code from a description
explain_code
Line-by-line explanations
review_code
Bugs, security, performance review
optimize_code
Speed, memory, readability
fix_error
Root cause and exact fix
explain_error
Plain-language error help
verify_code
Best-practices check
convert_code
Translate between languages
write_tests
Generate a full test file
security_audit
Severity-ranked vulnerability audit
build_architecture
Design a system architecture
solve
End-to-end coding task solver
deep_search
Typed research: web, reddit, github, news, academic, stats
ddg_search
Free DuckDuckGo links, no key
web_scrape
URL to Markdown, SSRF-guarded
get_ui_reference
UI/UX design references
content
Summarize, write, brainstorm, analyze, extract
compare
Side-by-side comparison
debate
Providers argue different stances
verify
Cross-check a claim with a confidence rating
crew
Role-based multi-model pipeline
run_workflow
Ordered multi-step pipeline
run_loop
Generate, review, improve until converged
proxima_cost_report
Token usage and estimated savings
proxima_agentic_status
Router, memory and health diagnostics
analyze_file
File or whole-folder analysis, secret scan
review_code_file
Review a single file on disk
clear_cache
Drop cached provider responses
show_window
Show the Proxima window
hide_window
Hide the window, MCP keeps running
toggle_window
Toggle window visibility
set_headless_mode
Run in the background
Simple setup, powerful results
Download the installer from GitHub releases (Windows installer; macOS and Linux builds are available too), or run from source with npm install and npm start.
Open Proxima and sign in to your AI accounts (ChatGPT, Claude, etc.)
Go to Settings and copy the MCP configuration JSON.
{
"mcpServers": {
"proxima": {
"command": "node",
"args": ["path/to/proxima/src/mcp/index.js"]
}
}
}
Paste config into your AI coding app (Cursor, VS Code, etc.) and enjoy!
Proxima unifies your existing accounts. Switch models instantly. Build faster.