MCP server exposing 7 tools for herb-tcm.
This URL is a JSON-RPC 2.0 endpoint over HTTP. Issue POST requests with a JSON-RPC body. Browsers and search crawlers land here on GET.
POST https://gateway.pipeworx.io/herb-tcm/mcp
Content-Type: application/json
{"jsonrpc":"2.0","id":1,"method":"tools/list"}herb_search — Search HERB 2.0 for a herb, ingredient, gene target or disease by Chinese name, pinyin, English name, Latin name, gene alias or HERB id — e.g. "板蓝根", "banlangen", "Dyers Woad" and "BAN LAN GEN" all resolve the same herb record. Returns matching ids to pass to herb_detail. Does not itself carry evidence tiers (it is a name lookup, not a relationship).herb_browse — Page through HERB 2.0's full herb, ingredient, target or disease list (7,263 herbs / 49,258 ingredients / 12,933 targets / 28,212 diseases). Use herb_search instead when you already have a name to look up.herb_detail — Full record for one HERB id: herb (composition + traditional-use summary + predicted and literature target/disease links), ingredient (structure + predicted and literature target/disease links), target (curated disease associations + which herbs are statistically linked to it), or disease (which targets and herbs are statistically linked to it). Every relationship row carries an evidence_tier. computational_prediction means a statistical or database-mined association with NO clinical or experimental confirmation — it is a hypothesis, not proof the herb/ingredient treats the condition. traditional_use reflects historical TCM practice, not a trial. Only human_clinical rows come from a paper that studied humans. A predicted herb-disease or herb-target edge (evidence_tier computational_prediction) is a screening hit from expression-overlap statistics, not proof of efficacy — never present it as "HERB shows herb X treats disease Y" without saying it is a prediction. Relationship tables are paged: each comes back as {total, offset, limit, returned, truncated, rows} with the TRUE upstream row count in `total` (a well-studied herb can have thousands of predicted disease rows — 25 are returned by default). truncated:true means more rows exist; page with offset, or narrow with `sections`.herb_papers — List PubMed-cited papers in HERB's reference index — the literature backing herb/ingredient-target and -disease associations. Each row is tagged evidence_tier human_clinical (its "Experiment type" includes "Clinical Experiment") or laboratory (cell/animal studies only). Filter by drug type or experiment type, or leave both blank to page through all ~2,000 references.herb_paper_detail — One reference's bibliographic record (journal, PubMed id, experiment type, phenotype) plus the specific targets/diseases that paper reports, tagged evidence_tier human_clinical or laboratory from the paper's own experiment type. Use the Paper id from a drug_paper_target/drug_paper_disease row (herb_detail) or from herb_papers.herb_experiments — List HERB's transcriptomic experiments (herb/ingredient treatment vs. control, GEO-deposited) — the raw evidence behind the differential-expression predictions in herb_experiment_detail. Filter by drug type, species or platform.herb_experiment_detail — Differential-expression results for one HERB transcriptomic experiment: top up/down-regulated genes and enriched GO terms / KEGG pathways, plus connectivity-map hit counts (compounds/knockdowns/overexpression signatures matching the expression pattern). Every row is evidence_tier computational_prediction — this is a statistical enrichment over one dataset, not a validated mechanism or a clinical finding.Code samples (curl / TypeScript / one-click client install), schemas, and the live playground are on the pack page:
https://pipeworx.io/packs/herb-tcm/
Pipeworx is an open MCP gateway connecting AI agents to live data. pipeworx.io