Engineering
mcpgraph avatar

mcpgraph

Build no-code MCP servers that orchestrate tools as directed graphs using YAML for data transformation, conditional routing, and automated workflows.

Introduction

mcpGraph is a powerful declarative framework for the Model Context Protocol (MCP) that enables users to create sophisticated agentic workflows without writing custom code. By defining MCP tools as directed graphs of nodes in YAML, it allows for the seamless orchestration of multiple MCP servers, complex data transformations via JSONata, and intelligent branching logic using JSON Logic. This solution is designed for engineers and developers looking to compose modular tools into high-performance, observable agent backends while maintaining strict security and traceability. It serves as an accessible alternative to traditional code-based agent development by providing a visualizable, declarative structure that remains easy to test, audit, and debug.

  • Orchestrate multiple MCP tools in sequence to handle multi-step tasks autonomously.

  • Implement data transformation pipelines between tool outputs and inputs using JSONata expressions.

  • Define complex routing logic with JSON Logic to handle conditional branching and decision-making.

  • Ensure complete observability by tracing every node execution, including inputs, outputs, and intermediate states.

  • Maintain standard-based configurations that are compatible with existing MCP clients like Claude Desktop.

  • Utilize a modular, declarative approach to build complex agent capabilities from simpler, reusable tool definitions.

  • Configuration is handled entirely through YAML, allowing version control and repeatability.

  • The execution engine supports custom execution limits such as max node iterations and execution timeouts to prevent runaway processes.

  • Standard outputs include structured execution history, timing data, and comprehensive logs for debugging purposes.

  • Requires local installation of the mcpGraph command-line tool or integration via standard MCP client configuration.

  • Intended for developers building LLM-based agents that require multi-tool interaction, data parsing, and logic-heavy task automation.

  • Integration is seamless with existing MCP ecosystem components, including filesystem and various API-based servers.

Repository Stats

Stars
15
Forks
2
Open Issues
0
Language
TypeScript
Default Branch
main
Sync Status
Idle
Last Synced
May 3, 2026, 10:26 PM
View on GitHub