Coala: a standard-based framework for converting CWL-described command-line tools into agentic toolsets.

Abstract

Large Language Models (LLM) can orchestrate computational analyses through agentic systems, but scaling their toolsets remains a barrier because tool definitions are often hard coded into agent implementation. We developed Command-line LLM-agent Adapter (Coala), a standards-based framework that bridges the Model Context Protocol (MCP) and the Common Workflow Language (CWL). Coala turns CWL tool descriptions into MCP-compatible, LLM-accessible schemas, treating tool definitions as data rather than code. Tools are then executed in containerized environments through a generic MCP server, which separates the agent's reasoning from tool execution. This framework improves reproducibility, reduces ongoing maintenance burden, and enables interactive access to local command-line tools through natural-language queries. Coala is available at https://coala.info and is openly developed on GitHub: https://github.com/coala-info/coala. Supplementary data are available at Bioinformatics online.