Pctx by Portofcontext is a free MCP server tool built with Rust, designed as an execution layer for AI agentic tool calls. It enables AI agents to interact with external systems by executing dynamically registered tools within an isolated Deno sandbox, running TypeScript code. Pctx aggregates multiple upstream MCP servers, providing a unified interface for agent interactions. Its core functionality focuses on secure and efficient execution of agent-generated code, making it a central component for advanced AI workflows.

What does Pctx connect to?

Pctx functions as an MCP server, connecting to various AI agents as clients. It also aggregates multiple upstream MCP servers, acting as a unified interface. This structure positions Pctx as an intermediary, orchestrating agentic tool calls and managing the execution layer. It supports interaction with AI clients like Claude, utilizing Programmatic Tool Calling, and OpenAI models through their function calling capabilities, establishing a versatile connection hub.

Core features and tools

Pctx offers a unified Code Mode interface, exposing tools as code functions. These tools are dynamically registered from upstream MCP servers, not a fixed set. AI agents can call these functions within a sandboxed environment. An example of a callable function is gdrive.getSheet. It supports HTTP as a stateless server for Code Mode sessions and STDIO for local MCP servers, ensuring flexible deployment and interaction methods.

Who is it for?

Pctx is designed for AI agents, developers crafting agentic workflows, and platform engineers managing external tools and MCP servers. It provides a crucial execution layer for complex AI interactions. This tool is valuable for those needing to orchestrate multi-step processes or handle large datasets efficiently with AI. Skip this if you do not require an execution layer for AI agents to interact with dynamically registered external tools.

How to get started

Access is granted through a standard OAuth authorization flow for HTTP-based transports, following OAuth 2.1 conventions. Pctx handles authentication by configuring "pctx.json" to source secrets from environment variables, the system keychain, or external commands.

Use cases

• Orchestrating multi-step workflows with three or more dependent tool calls by allowing AI agents to write code that executes in a sandbox, reducing latency and token consumption.
• Processing large datasets by enabling AI agents to filter, sort, or transform results within the sandbox before the final output is returned, thereby reducing data passed through the model's context window.
• Aggregating multiple upstream MCP servers into a single interface, providing persistent upstream sessions for efficient agent interaction.

Pctx provides a secure and efficient execution environment for AI agents, abstracting complex tool interactions into a unified Code Mode. It is particularly effective for managing multi-step workflows and large datasets, while aggregating diverse MCP servers. This platform helps developers and platform engineers build sophisticated agentic systems.