nf-core MCP server

A toy MCP server that lets an LLM agent find, validate, launch and monitor nf-core pipelines

nf-core MCP server is a toy project that connects a large language model (LLM) to nf-core pipelines through the Model Context Protocol (MCP). With tool calling, an AI agent such as Claude can take a researcher from a data problem to a running analysis: find the right pipeline, build its samplesheet, check it, launch the run and follow it.

I built it to find out how far an agentic workflow gets with pipelines I know from the inside. The server exposes the steps a bioinformatician would take as tools the model can call:

1. Search: a BM25 index over the nf-core catalog and pipeline READMEs, to match a research question to a pipeline.
2. Read: the pipeline’s documentation and its parameter schema at a pinned release.
3. Prepare: a samplesheet generated from a folder of FASTQ files.
4. Validate: samplesheet and parameters checked against the pipeline’s own nextflow_schema.json and input schema.
5. Run: nextflow run in the background, followed with status and error reports, and summarised with QC numbers, software versions and a methods text.