Awesome MCP › Agent Orchestration Platforms

Chen-zexi/open-ptc-agent

⭐ 730 Python added to this list on 2025-12-15 repository created 2025-11-07

Open PTC Agent is an open source project implementing the Programmatic Tool Calling (PTC) paradigm introduced by Anthropic. PTC allows large language models (LLMs) to invoke tools through code execution rather than individual JSON tool calls, enabling more efficient and powerful workflows. This approach leverages LLMs' strengths in writing code, understanding context, and reasoning about data flows to orchestrate complex tasks. By running code in a sandbox environment, PTC significantly reduces token usage by processing large volumes of structured data locally and returning only final outputs to the model. This is especially beneficial for handling time series data, financial market data, and scenarios requiring data filtering, aggregation, transformation, or visualization. The project is built on LangChain DeepAgents and uses the Daytona sandbox for secure code execution with filesystem isolation and snapshot support. It features an interactive CLI for terminal-based interaction with session persistence, plan mode, themes, and a rich user interface. Background subagent execution allows asynchronous task handling with automatic result collection, enhancing agent control and efficiency. The agent supports vision and multimodal capabilities, enabling analysis of images from various sources. Open PTC Agent supports universal MCP by auto-converting any MCP server tools to Python functions and offers progressive tool discovery to avoid upfront token overload. It includes enhanced file operation tools optimized for sandbox environments and supports auto image upload to cloud storage providers like Cloudflare R2, AWS S3, and Alibaba OSS. The project is compatible with LangGraph Cloud/Studio and supports multiple LLM providers including Anthropic and OpenAI. The architecture includes core agent libraries, an interactive CLI application, demo MCP server implementations, and example notebooks and scripts. Native tools cover code execution, shell commands, file operations, and content search, while middleware provides subagent delegation, asynchronous execution, image viewing, task planning, and conversation summarization. The agent's design facilitates complex multi-step tasks with background subagents running asynchronously, making it a powerful tool for advanced programmatic tool calling workflows.

https://github.com/Chen-zexi/open-ptc-agent

agentasynchronous-tasksbackground-subagent-executioncloud-storagecode-executiondaytonadaytona-sandboxfile-operationsfinancial-data-analysisinteractive-clilangchainlangchain-deepagentslanggraphlangraphllmllm-providersmcpmultimodal-llmsprogrammatic-tool-callingptcpythonsandbox-environmentsubagentstask-monitoringtool-discoveryvision-support