Your AI agent's GPU compute engine.

MCP Server

7 structured tools via JSON-RPC 2.0. Add one line to your config and your AI agent gets GPU compute, statistics, ML, and data analysis. No API keys. No cloud. Everything local.

Tools

7 tools. One binary.

Each tool accepts structured JSON input and returns structured JSON output. Your agent calls them like any other MCP tool.

octoflow_run

Execute Code

Run OctoFlow code directly. GPU operations dispatch to Vulkan compute shaders.

{"code": "let a = gpu_fill(1.0, 1000000)\nprint(\"Sum: {gpu_sum(a)}\")"}
octoflow_chat

Natural Language

Describe what you want in English. Built-in Qwen3-1.7B generates, runs, and auto-fixes code.

{"prompt": "sort 1M random numbers on GPU and show the top 10"}
octoflow_check

Validate Syntax

Check .flow code for errors without running it. Returns structured diagnostics.

{"code": "let x = gpu_fill(1.0, 100)\nprint(\"Done: {x}\")"}
octoflow_gpu_sort

GPU Sort

GPU-accelerated sorting. Pass data, get sorted results. Millions of elements in milliseconds.

{"data": [3.1, 1.4, 2.7, ...], "descending": false}
octoflow_gpu_stats

GPU Statistics

Mean, stddev, correlation, SMA, EMA, Sharpe ratio. All computed on GPU.

{"data": [12.5, 14.2, ...], "operations": ["mean", "stddev", "sharpe"]}
octoflow_image

Image Processing

Load, transform, and save images. Blur, resize, crop, convert. BMP, GIF, PNG support.

{"op": "blur", "path": "photo.bmp", "kernel": 5}
octoflow_csv

CSV Analysis

Load a CSV and run queries. Statistics, filtering, aggregation, correlation, trend analysis.

{"path": "sales.csv", "query": "mean, median, stddev, trend"}
Setup

One line. That's it.

Add this to your AI assistant's MCP configuration. OctoFlow starts as a server and your agent discovers all 7 tools automatically.

settings.json / openclaw.json / .cursor/mcp.json
// Add to your MCP server configuration
{
  "mcpServers": {
    "octoflow": {
      "command": "octoflow",
      "args": ["mcp-serve"]
    }
  }
}

// That's it. Your agent now has GPU compute.
Compatibility

Works with your stack.

Any MCP client that supports stdio transport can use OctoFlow as a tool provider.

OpenClaw

Add to ~/.openclaw/settings.json or install as a skill via npx clawhub install octoflow. 220K+ users. The agent automatically suggests OctoFlow for GPU tasks.

Claude Desktop

Add to claude_desktop_config.json. Claude discovers all 7 tools and uses them for data analysis, GPU compute, and ML tasks.

Cursor

Add to .cursor/mcp.json in your project. The AI coding assistant gains GPU compute, statistics, and image processing capabilities.

Any MCP Client

OctoFlow speaks JSON-RPC 2.0 over stdio. Any tool that supports the Model Context Protocol can use it. Windsurf, Cline, custom agents — all work.

Recipes

Common agent workflows.

7 built-in recipe scripts in stdlib/agent/ for the most common agent tasks.

Data Analysis

Agent sends CSV, gets a full statistical report. Mean, median, stddev, correlation, trend, outliers.

Time Series

SMA, EMA, RSI, MACD, Bollinger bands. GPU-parallel across multiple tickers. Alert on signal alignment.

GPU Sort

Sort millions of elements on GPU in milliseconds. Returns sorted data with timing metrics.

Hypothesis Testing

One-sample and two-sample t-tests, chi-squared, ANOVA. Returns test statistic, p-value, significance.

Classification

K-means clustering and KNN classification. Agent sends data, gets cluster assignments or predictions.

LLM Translate

Use OctoFlow's built-in Qwen3-1.7B for code generation, translation, and summarization tasks.

7
MCP Tools
7
Agent Recipes
JSON-RPC
2.0 Protocol
0
API Keys Needed
4.5 MB
Total Size

Give your agent GPU compute.

4.5 MB download. Zero dependencies. Windows, Linux, macOS.

Download v1.5.9 Documentation