Bold Systems MCP server

@pipeworx/bold-systems

The global DNA barcode reference library from BOLD Systems (Barcode of Life Data System, Centre for Biodiversity Genomics, University of Guelph) — specimen vouchers with locality, institution and taxonomy, the COI/rbcL/matK barcode sequences themselves, and BIN species-proxy clusters.

Part of Pipeworx — an MCP gateway connecting AI agents to 1679+ live data sources.

Tools

process ID, full taxonomy, BIN, collection date, locality with coordinates, holding institution, marker and GenBank accession. Filters AND together.

• bold_specimens(taxon?, country?, bin?, institution?, query?, limit?, offset?) — specimen records:

barcode sequences, with primers, base count, INSDC accession and a ready-to-paste FASTA block.

• bold_sequences(taxon?, country?, bin?, institution?, marker?, query?, limit?, offset?) — the

counts at each rank above it, a plain-language description, and alternative matches with counts.

• bold_taxonomy(name, rank?) — resolve a name against BOLD's own index: lineage, record/species/BIN

Auth

Keyless.

Data sources

, NOT under /api/).

• — term index; resolves a plain name to a query triplet.
• — turns a triplet query into a query_id.
• — the records, nuc sequence inline.
• — lineage, counts, prose.
• Swagger: (the spec itself is at

Traps

The v4 API is dead. v4.boldsystems.org/index.php/API_Public/{specimen,sequence,combined} — what every tutorial and most R packages still point at — answers with a "BOLD Public Offline" HTML page, not JSON, so a naive client reports a parse error rather than a retirement. Everything here is the v5 portal API.

The query language is scope:field:value triplets and the scope/field pair is not guessable from the value. Canada is geo:country/ocean, Danaus is tax:genus, BOLD:AAA9566 is bin:uri, Smithsonian Institution is inst:name. Free text gets a 400 {"detail":"Invalid triplet token"}. The pack resolves every typed argument through /api/terms first, which is what turns a plain name into a query that matches instead of a 400 or a silent zero.

/api/counts reports the SUM of the per-term counts, not the size of the intersection. For tax:species:Danaus plexippus;geo:country/ocean:Canada it says 2,520,553 records; the actual match set is 7. That is a clean 200 carrying a number wrong by five orders of magnitude, and it is the number a naive integration would quote. The pack does not use it — matchingRecords comes from recordsTotal on the document query, which is the real count. (Semicolon-separated triplets AND on /documents even though they sum on /counts.)

/api/query/preprocessor falls back to ID-field matching on a multi-word string. Feeding it "Danaus plexippus Canada" returns ids:processid:...;ids:sampleid:...;ids:insdcacs:... — a query that matches nothing and errors nowhere. Resolve each filter separately, which is what resolveTriplet does.

BOLD holds specimen records that were never sequenced. bold_sequences filters to records with a nuc field and says so in note when the result is empty, so "this taxon has no barcodes" is distinguishable from "this taxon has no records".

The marker filter is applied to the returned page, not upstream (BOLD's term index has no entry for COI-5P). The pack over-fetches 5× when a marker is set and reports which markers were actually present in the page it scanned, so an empty result names the markers you could have asked for.

Shortened. The full README is on GitHub.