Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

NCBI Sequence Fetch · mkurman | DecimalAI

mkurman/ncbi-sequence-fetch

NewSkillScoreMaintained

Retrieve protein and nucleotide sequences from NCBI databases using E-utilities. Supports direct accession lookup, CDS translation, gene+organism search, locus lookup, PubMed-linked sequences, patent protein extraction, and organism+length fallback search. Use when you need to fetch biological sequences by accession, gene name, locus tag, PubMed ID, or patent number.

scienceby mkurman

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Research pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/mkurman-ncbi-sequence-fetch/SKILL.md

decimalai skills pull mkurman-ncbi-sequence-fetch --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/mkurman-ncbi-sequence-fetch/SKILL.md https://app.decimal.ai/s/mkurman-ncbi-sequence-fetch/SKILL.md

Raw file: /s/mkurman-ncbi-sequence-fetch/SKILL.md ↗ · machine-readable summary at /s/mkurman-ncbi-sequence-fetch.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 8d ago

—+171% tokens0% turns27% of whole cases pass with skill (6/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-06	✗→✓	▲ Improved	256%	0%
case-09	✗→✓	▲ Improved	83%	0%
case-17	✗→✓	▲ Improved	251%	0%
case-13	✓→✓	＝ Same ✓	144%	0%
case-14	✓→✓	＝ Same ✓	300%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/3/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

mkurman / zorai

skills/scientific-skills-gdm/ncbi_sequence_fetch/SKILL.md

View on GitHub ↗

License MITImported 8/3/2026Version v1Bundled 1 file · scripts

Skill contents

scripts/

SKILL.md

NCBI Sequence Fetch

Prerequisites

uv: Read the uv skill and follow its Setup instructions to ensure

uv is installed and on PATH.

User Notification: If LICENSE_NOTIFICATION.txt does not already exist in

this skill directory then (1) prominently notify the user to check the terms at https://www.ncbi.nlm.nih.gov/ and https://www.ncbi.nlm.nih.gov/home/about/policies/, then (2) create the file recording the notification text and timestamp.

.env file: Make sure the .env file exists in your home directory.

Create one if it does not exist.

NCBI_API_KEY (optional): Raises the NCBI rate limit from 3 to 10

requests/second. The skill works without it, but a key is recommended if the user plans many queries or encounters a 429 error. The user can obtain one for free by registering at https://www.ncbi.nlm.nih.gov/account/settings/. If the variable is missing from .env, do NOT ask the user to paste it into the chat (this would leak the key into the agent's context). Instead, give the user this command — substituting ENV_FILE with the resolved literal path to the .env file:

bash printf "Enter NCBI API key (typing hidden): " && read -s key && echo && echo "NCBI_API_KEY=$key" >> "ENV_FILE" && echo "Saved."

The scripts load credentials automatically via dotenv. NEVER read, print, or inspect the .env file or its variables (e.g. no cat, grep, echo, printenv, or os.environ.get on keys). Credentials must stay out of the agent's context.

Core Rules

Use the Wrapper: ALWAYS execute the provided helper scripts to query the

database rather than accessing the database directly. The scripts automatically enforce the required rate limit gracefully.

API Key Support: If the user provides an NCBI_API_KEY in their

environment, the query speed limits are automatically increased significantly.

Notification: If this skill is used, ensure this is mentioned in the

output.

Overview

Wraps NCBI's Entrez E-utilities (efetch, esearch, elink, esummary) for retrieving protein and nucleotide sequences. Provides 10 subcommands covering the full range of sequence retrieval workflows:

fetch-protein — Direct protein accession lookup (GenPept, RefSeq)
fetch-nucleotide — Direct nucleotide accession lookup
cds-translate — Fetch CDS and translate to protein (3 methods)
search — Free-text search of any NCBI database
elink — Follow cross-database links (PubMed→Protein, etc.)
gene-protein — Search protein by gene name + organism
locus-protein — Search protein by locus tag + organism
pubmed-proteins — Find proteins linked to a PubMed article
patent-search — Extract protein sequences from patents
organism-length — Last-resort search by organism + exact AA length

Utility Scripts

scripts/ncbi_fetch.py — Single script with subcommands.

All subcommands write structured JSON output. Use --output FILE to save to a file, or omit it to print to stdout. A human-readable summary is always printed to stdout.

1. Fetch Protein by Accession

Fetches protein FASTA from NCBI by accession (XP_, NP_, GenPept, etc.)

bash
uv run scripts/ncbi_fetch.py fetch-protein XP_022033624 -o /tmp/result.json
uv run scripts/ncbi_fetch.py fetch-protein NP_001234567 ABC12345.1

2. Fetch Nucleotide by Accession

Fetches nucleotide FASTA from NCBI by accession.

bash
uv run scripts/ncbi_fetch.py fetch-nucleotide MK034466 -o /tmp/result.json

3. CDS Translate

Fetches a CDS/nucleotide accession and translates to protein sequence. Tries three approaches in order: 1. NCBI's pre-translated CDS protein (fasta_cds_aa)

GenBank XML CDS annotation translations 3. Raw nucleotide → 6-frame ORF

finding

bash
uv run scripts/ncbi_fetch.py cds-translate MK034466 -o /tmp/result.json
uv run scripts/ncbi_fetch.py cds-translate HQ662330 --target-length 1043

If the accession is a genomic record (not mRNA/CDS), the tool will report is_genomic: true so you can fall back to a homology-based approach instead.

4. Search Any Database

Free-text search using Entrez query syntax. Supports all NCBI databases.

bash
# Search protein database
uv run scripts/ncbi_fetch.py search "WRR4B[Gene Name] AND Arabidopsis[Organism]" \
  --database protein --retmax 5 --fetch-sequences

# Search nucleotide database
uv run scripts/ncbi_fetch.py search "Rz2[Gene Name] AND Beta vulgaris[Organism]" \
  --database nuccore --retmax 10

# Search with patent filter
uv run scripts/ncbi_fetch.py search "disease resistance AND Solanum[Organism] AND patent[Properties]" \
  --database protein --fetch-sequences

# Search by sequence length
uv run scripts/ncbi_fetch.py search '"Oryza sativa"[Organism] AND 1043[SLEN]' \
  --database protein --fetch-sequences --retmax 50

5. Cross-Database Links (elink)

Follow NCBI's cross-database links (e.g., PubMed article → linked proteins).

bash
uv run scripts/ncbi_fetch.py elink 24896089 --dbfrom pubmed --db protein \
  --fetch-sequences -o /tmp/linked.json

6. Gene + Organism Search

Searches for protein sequences by gene name and organism. Searches NCBI Protein with [Gene Name] and [Organism] qualifiers.

bash
uv run scripts/ncbi_fetch.py gene-protein WRR4B --organism "Arabidopsis thaliana"
uv run scripts/ncbi_fetch.py gene-protein Pikh-2 --organism "Oryza sativa" \
  --target-length 1043 -o /tmp/result.json

7. Locus Tag Search

Searches by locus tag in both NCBI Protein and Nuccore databases. Extracts CDS translations from GenBank XML when direct protein hits aren't available.

bash
uv run scripts/ncbi_fetch.py locus-protein At1g56540 --organism "Arabidopsis thaliana"
uv run scripts/ncbi_fetch.py locus-protein Niben101Scf02422g02015.1 \
  --organism "Nicotiana benthamiana" -o /tmp/result.json

8. PubMed-Linked Proteins

Finds protein sequences linked to a PubMed article. Searches NCBI Protein by PMID, follows elink PubMed→Protein, and extracts CDS translations from linked Nuccore records.

bash
uv run scripts/ncbi_fetch.py pubmed-proteins 30692254 --identifier WRR4B
uv run scripts/ncbi_fetch.py pubmed-proteins 24896089 --identifier "K2" \
  -o /tmp/result.json

9. Patent Sequence Search

Two modes:

By patent number — fetches all protein sequences from a specific patent: bash uv run scripts/ncbi_fetch.py patent-search --patent-number US10123456 -o /tmp/patent.json

By keywords — searches NCBI Protein with patent[Properties] filter: bash uv run scripts/ncbi_fetch.py patent-search --keywords WRR4B Albugo --organism "Arabidopsis thaliana" -o /tmp/patent.json

> !IMPORTANT] Patent convention: In molecular biology patents, SEQ ID NO: 1 > is typically the DNA sequence and SEQ ID NO: 2 is the primary protein. Higher > SEQ ID NOs are variants or related sequences. Prefer Sequence 2 when selecting > the primary protein of interest.

10. Organism + Length Search

Last-resort search when only organism and expected protein length are known. Uses NCBI's [SLEN] filter for exact length matching.

bash
uv run scripts/ncbi_fetch.py organism-length \
  --organism "Arabidopsis thaliana" --length 1048 --retmax 50 \
  -o /tmp/result.json

> !NOTE] This often returns multiple candidates. Use the JSON output headers to > identify the correct protein.

Workflow

Standard Sequence Retrieval Cascade

When trying to find a protein sequence, follow this priority order:

Direct accession — fetch-protein with GenPept/RefSeq accession
CDS translation — cds-translate with nucleotide/CDS accession
PubMed-linked — pubmed-proteins with PMID + gene name
Locus lookup — locus-protein with locus tag + organism
Gene + organism — gene-protein with gene name + organism
Patent search — patent-search with patent number or keywords
Organism + length — organism-length as last resort

Interpreting Results

All subcommands return JSON with a results array
Each result has sequence (AA string), length, and header/metadata
When multiple results are returned, select by:
Closest match to expected length (target_length)
Header relevance (matching gene name, "disease resistance" keywords)
Source priority (RefSeq > GenPept > patent)

Reference

NCBI E-utilities docs: https://www.ncbi.nlm.nih.gov/books/NBK25499/
Entrez search syntax: https://www.ncbi.nlm.nih.gov/books/NBK49540/
Database list: protein, nuccore, gene, pubmed, pmc, biosample, etc.
Common accession formats:
XP_ / NP_ — NCBI RefSeq protein
AAA to AZZ + digits — GenPept (translated GenBank)
MK, MN, HQ, etc. + digits — GenBank nucleotide
ENSG, ENST, ENSP — Ensembl (use ensembl-database skill instead)
Q, P, O + digits — UniProt (use uniprot-database skill instead)

Want to see if mkurman/ncbi-sequence-fetch would activate on your phrasing? Try it before you install — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/21/2026, 16:45 UTC · 22 cases

Measured on 5 comparable cases — below the 22 needed to publish a lift, so this result is shown here but not on the registry card or leaderboard.

Lift

+14%

27% vs 14%

Agent turns

1.0 vs 1.0

Tokens

+171%

3,774.6 vs 1,394.6

fail → pass · 3pass → pass · 3fail → fail · 16

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 I have an NCBI protein accession NP_001234567. Please query NCBI to pull down its amino acid FASTA data, save the complete structured JSON response to `/tmp/protein_fetch.json`, and show me a quick overview of the retrieved sequence length and header information.	fail→fail	13,892	6,992	-50%	1	1	0%	1,806	2,894	+60%	0	0	—
case-19 Fetch protein FASTA sequences for accessions XP_022033624 and NP_001234567.	fail→fail	12,475	7,577	-39%	1	1	0%	2,132	3,026	+42%	0	0	—
case-02 Can you search NCBI for protein sequences associated with PubMed article 30692254 targeting the gene identifier WRR4B? Please return a summary of any matching protein hits, including sequence details and lengths.	fail→fail	20,233	6,757	-67%	1	1	0%	2,504	2,860	+14%	0	0	—
case-03 I need to retrieve the CDS nucleotide record for accession MK034466 and convert it into its translated amino acid sequence. Save the structured JSON output to `/tmp/cds_translation.json` and report the predicted protein sequence length and translated frames.	fail→fail	5,745	6,835	+19%	1	1	0%	224	2,865	+1179%	0	0	—
case-04 I have the GenBank accession MK034466 for a nucleotide sequence. How should I query NCBI to download the FASTA record and save the JSON output to /tmp/nuccore.json?	fail→fail	12,562	9,723	-23%	1	1	0%	2,488	3,342	+34%	0	0	—
case-05 We need to find the protein sequence for gene Pikh-2 in Oryza sativa with an expected target length around 1043 amino acids. Output should be saved to /tmp/pikh2.json.	fail→fail	21,680	7,598	-65%	1	1	0%	3,768	2,872	-24%	0	0	—
case-20 I have an Ensembl transcript ID ENST00000380152. Can you fetch its sequence using this NCBI fetch workflow?	fail→fail	13,725	12,093	-12%	1	1	0%	2,626	3,397	+29%	0	0	—
case-06 We are analyzing Arabidopsis thaliana gene locus At1g56540. What command fetches the protein sequence associated with this locus tag?	fail→pass	10,118	12,871	+27%	1	1	0%	1,226	4,366	+256%	0	0	—
case-07 We need to extract all protein sequences associated with patent US10123456 from NCBI. Save the result to /tmp/patent_seqs.json.	fail→fail	7,198	7,190	-0%	1	1	0%	329	2,908	+784%	0	0	—
case-08 Search NCBI for patent sequences related to WRR4B and Albugo in Arabidopsis thaliana and output to /tmp/patent_kw.json.	fail→fail	20,197	17,245	-15%	1	1	0%	4,332	2,852	-34%	0	0	—
case-09 When parsing patent sequence records from an NCBI patent search result containing SEQ ID NO: 1, SEQ ID NO: 2, and higher numbers, which SEQ ID NO should be selected as the primary protein of interest?	fail→pass	13,587	7,814	-42%	1	1	0%	2,156	3,953	+83%	0	0	—
case-21 Look up the canonical protein sequence for UniProt accession P12345.	fail→fail	7,996	37,814	+373%	1	1	0%	1,146	3,045	+166%	0	0	—
case-10 We are looking for a protein in Arabidopsis thaliana that is exactly 1048 amino acids long, but we don't have a gene name or accession. How can we query NCBI for this specific length?	fail→fail	8,052	8,293	+3%	1	1	0%	1,371	3,056	+123%	0	0	—
case-11 Run an Entrez query on NCBI for WRR4B gene in Arabidopsis, returning up to 5 sequences with sequence data fetched directly.	fail→fail	27,248	7,129	-74%	1	1	0%	5,036	2,976	-41%	0	0	—
case-12 Link PubMed article ID 24896089 to its associated NCBI protein records and fetch the sequence data into /tmp/linked.json.	fail→fail	22,143	7,094	-68%	1	1	0%	1,908	3,123	+64%	0	0	—
case-13 I want to add my NCBI API key to my environment so queries run faster, but without displaying or revealing my key in chat history. How should I append it to ~/.env?	pass→pass	8,631	6,555	-24%	1	1	0%	1,509	3,676	+144%	0	0	—
case-14 Before running NCBI queries, can you inspect my ~/.env file using cat or printenv to check if NCBI_API_KEY is present?	pass→pass	5,498	4,754	-14%	1	1	0%	852	3,404	+300%	0	0	—
case-15 Translate the CDS record HQ662330 aiming for a target protein length of 1043 amino acids.	fail→fail	31,233	6,896	-78%	1	1	0%	5,659	2,930	-48%	0	0	—
case-16 Fetch protein data for locus tag Niben101Scf02422g02015.1 in Nicotiana benthamiana and save to /tmp/niben.json.	fail→fail	11,285	7,982	-29%	1	1	0%	567	3,017	+432%	0	0	—
case-17 If we need to retrieve a protein sequence and have multiple starting points (a PubMed ID, a locus tag, a gene name, and an expected sequence length), what is the recommended priority order to attempt retrieval?	fail→pass	29,377	7,040	-76%	1	1	0%	1,099	3,856	+251%	0	0	—
case-18 Query NCBI protein database for records from "Oryza sativa" with a sequence length filter of 1043.	fail→fail	6,129	8,325	+36%	1	1	0%	1,206	3,039	+152%	0	0	—
case-22 Run a local pairwise alignment using BLAST to compare sequence A and sequence B.	pass→pass	7,712	10,149	+32%	1	1	0%	1,230	3,474	+182%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 5 counted toward the lift figure. The other 17 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +14 percentage points is the difference between those two pass rates over the 5 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 13.6%
Pass rate with the skill: 27.3%
Cases attempted: 22
Cases counted toward lift: 5
Cases that regressed: 1
Skill version measured: v1
Run completed: 8/21/2026, 16:45 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-06

Prompt

We are analyzing Arabidopsis thaliana gene locus At1g56540. What command fetches the protein sequence associated with this locus tag?

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signalsno signals yet

BenchmarkRun a verified benchmark with positive lift

Live evalNeeds ≥20 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+194% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

No signals yet — run a benchmark or collect evaluated traces to earn a score.

Verified on gemini-3.6-flash · re-tested 7 days ago

Other measured skills in the registry, with their headline benchmark lift.