Genoplot
Server Details
Ancestry on G25 and raw DNA: nMonte fits, PCAs, GEDmatch-style and G25 calculators, qpAdm models.
- Status
- Healthy
- Uptime
- 4.3% over the last 24 hours
- OAuth
- Works in Glama
- Last Tested
- Transport
- Streamable HTTP · MCP 2025-11-25
- URL
- ChatGPT
- –
- Claude
- –
TDQS
Scored across 20 tools
Most tools target clearly distinct resources or workflow stages, and the descriptions explicitly separate similar pairs such as add_raw_file/get_raw_upload and list_history/get_saved_result. The main ambiguity is between run_calculator (G25 coordinates) and run_raw_calculator (raw DNA file), plus the related distinction between computed and simulated G25 coordinates, but the descriptions provide enough cues to tell them apart.
The set uses consistent snake_case and is dominated by predictable verb_noun names such as add_raw_file, get_credit_balance, list_calculators, and run_qpadm. A few tools deviate into noun-phrase or domain-specific naming (purchase, purchase_status, nearest_populations, nmonte_fit), but they remain readable and do not create real confusion.
Twenty tools is somewhat heavy, but the server covers a broad genetic-genealogy platform with uploads, calculators, population search, nMonte, PCA, qpAdm, history, credits, purchases, and guides. Each tool maps to a distinct operation or workflow stage, so the count is mostly justified rather than bloat.
The surface covers the main lifecycle: raw upload, upload status, sample listing, G25 creation, calculator discovery and execution, population search/description, nMonte, PCA, qpAdm submit/poll, saved results, credits, and purchases. Minor gaps remain, such as deleting samples or results, accepting terms directly, and running other formal statistics like FST or f3, but these can reasonably be handled in the Genoplot app.
Available Tools
20 toolsadd_raw_fileAdd a raw DNA fileAInspect
Gives the user a private link for adding their raw DNA file (23andMe, Ancestry, FamilyTreeDNA (FTDNA), LivingDNA, MyHeritage, WeGene) to their Genoplot account. The file goes from the user's browser straight to Genoplot's encrypted storage and never passes through this conversation, so never ask the user to paste, attach or upload the file here, and never open the link yourself. Give the user the link: it works once, for 30 minutes, for one file up to 120 MB, and the screen asks for the testing company and a sample name (pass them if the user already said). Then ask the user to tell you when the screen says the file was added, and call get_raw_upload with upload_id; do not call it in a loop. Free; needs accepted terms. A whole-genome VCF is added in the app instead. How to read this: the link only adds a file to this user's account; it cannot read anything. run_raw_calculator runs GEDmatch-style calculators on it on any plan. With a Basic or Pro plan: make_g25_from_raw makes its G25 coordinates (one G25 set; annual plans include sets), which nmonte_fit, nearest_populations and run_calculator need; run_qpadm takes it as a target. get_credit_balance shows the plan and the G25 sets left. On the Free plan, tell the user what a plan adds and offer one with purchase (kind subscription).
Inputs: name, vendor.
describe_populationsDescribe reference populationsARead-onlyIdempotentInspect
Catalogue facts for named populations: id, sample count, modern or ancient, horizon, origin, date. Accepts any catalogue spelling ("Iranian_Fars", "Iranian Fars", "Iranian_Fars|22"). How to read this: an entry marked "not in the catalogue" is a name Genoplot does not have; search_populations finds the closest spellings.
Inputs: names (required).
get_credit_balanceCredit balance and pricesARead-onlyIdempotentInspect
The user's Genoplot credit balance, plan, and what each paid tool costs in credits, plus whether the account is ready for paid runs (terms accepted). Free. How to read this: credits are the same credits as in the app and refill with the subscription; students (Basic Student plan) spend purchased credits on runs through connected assistants, shown as paid_credits. nmonte_fit charges a fixed price per fit. run_calculator, run_raw_calculator and run_qpadm are charged by the run itself at Genoplot's standard rates, so their cost appears as the balance change after the run (run_raw_calculator reports it); a qpAdm submit requires the minimum balance shown. g25_sets is how many raw files make_g25_from_raw can still turn into G25 coordinates (a set each, not credits). When the balance is short, the purchase tool prepares a subscription checkout or buys credits.
Inputs: no named parameters.
get_qpadm_resultRead a qpAdm resultARead-onlyIdempotentInspect
The status and, when finished, the result of a qpAdm job submitted with run_qpadm (or from the app): the model's p-value, chi-squared and degrees of freedom, feasibility, each source's coefficient with its standard error and z-score, the outgroups, and the coverage of the target file. Free. Status "pending" means the job has not been written yet (poll again in 10 seconds; an id still unknown ten minutes after its submit, or one run_qpadm did not return, fails with not_found), "running" that it is computing, "success" that models are present, "error" that it failed (the message says why, including an insufficient-credits refusal). How to read this: the p-value is the probability of observing data like the user's if the model were true. It is not the probability that the model is true, and it is not a score of how good the model is. Above 0.05 the model is not rejected (the data are consistent with it); below 0.05 it is rejected (the sources cannot account for the target's relationships to the outgroups); a p-value near 1 often means the outgroups lack the power to distinguish anything, not that the truth was found. Not rejected is not the same as correct: several mutually exclusive models routinely pass for the same target and the statistics cannot choose between them. "feasible: false" (a negative coefficient) means the model is mathematically inadmissible even if its p-value is high. Standard errors are the honest width of each weight: 54% ± 10.5 means roughly 43 to 64%. Coverage matters: a consumer chip calls under a tenth of the reference SNPs, which widens the errors; a comfortable p-value on a low-coverage file is weaker evidence than the same p-value on a dense or imputed one, and with allsnps true p-values are anti-conservative when outgroups vary in coverage, so re-run a borderline model with allsnps false.
Inputs: jobid (required).
get_raw_uploadCheck a raw file uploadARead-onlyIdempotentInspect
Where an upload started with add_raw_file stands: waiting (the link has not been used yet), uploading or checking (in progress on the user's screen), invalid (the last file did not read as the chosen testing company's; the user can try again on the same screen), added (the sample reference is returned), or expired. Free. Call it when the user says the upload is done, not in a loop. How to read this: an added sample appears in list_my_samples with has_raw_file true. run_raw_calculator runs GEDmatch-style calculators on it on any plan. With a Basic or Pro plan: make_g25_from_raw makes its G25 coordinates (one G25 set; annual plans include sets), which nmonte_fit, nearest_populations and run_calculator need; run_qpadm takes it as a target. get_credit_balance shows the plan and the G25 sets left. On the Free plan, tell the user what a plan adds and offer one with purchase (kind subscription).
Inputs: upload_id (required).
get_saved_resultOpen a saved resultARead-onlyIdempotentInspect
Opens one saved result from list_history as data. Free. By kind: an nMonte fit gives each target's source percentages, distance and band, nearest populations and the run's settings; a calculator run each target's component percentages, fit distance and nearest populations; an Admix+ analysis its component percentages with the report's reading of each, the closest populations and the conclusion; a PCA plot its axes (with each axis's share of the variance), every plotted population's centre and spread, the user's and reference samples' positions, and with points: true every point; a PCA analysis the plot's facts and the report's sections; a formal-stats run its populations and its models (p-values, coefficients, standard errors) or statistics; a formal-stats analysis the best model, the models tested and the report. A row still in progress returns its state only. How to read this: these are the numbers as the user saved them, with the settings they were run with; a saved distance is comparable only to distances of the same target, coordinate type and settings, and the text of an AI report is that report's interpretation, not a new measurement. To test something further, run it: nmonte_fit, run_pca, run_qpadm.
Inputs: id (required), points.
list_calculatorsList G25 calculatorsARead-onlyIdempotentInspect
The public G25 admixture calculators Genoplot hosts, optionally filtered by words in the name or description: each one's id (what run_calculator takes), display name, creator, number of components and whether it runs on scaled coordinates. With kind "raw", the GEDmatch-style admixture calculators that run_raw_calculator runs on a raw DNA file instead (Eurogenes K13, Dodecad K12b, HarappaWorld …), with their components and the balance a run needs to start (a run costs about 20 to 30 credits). Free. How to read this: calculators differ in their components and reference sets; a calculator's name usually says what it is for (regional, global, ancient). Ask the user which they want rather than choosing for them when several fit.
Inputs: kind, limit, query.
list_historyList saved resultsARead-onlyIdempotentInspect
Lists the results the user saved in Genoplot's History, newest first: nMonte fits, calculator runs, Admix+ analyses (AI reports on a calculator or nMonte result), PCA plots and PCA analyses (AI reports on a plot), formal-stats runs (qpAdm, FST, f3) and formal-stats analyses. Free. Filter by kind (the History chips: nmonte also lists the Admix+ analyses of nMonte runs, pca and formal_stats also their AI reports), search by words in the name, a target, a source or the calculator, and page through. Each row carries an id for get_saved_result, which opens the result itself, its exact kind, and a state: saved, in_progress (a report still being written) or failed (runs that errored are left out unless include_failed). qpAdm jobs started here or in the app and not saved to History are read with get_qpadm_result. How to read this: a row is what the user saved and when; its targets and sources are labels, and the numbers are only in get_saved_result.
Inputs: kind, page, search, page_size, +1 more.
list_my_samplesList my saved samplesARead-onlyIdempotentInspect
The user's own samples saved on Genoplot: each one's reference (":", what the other tools accept as a target), alias, whether it is a RAW file upload (usable as a qpAdm target and by run_raw_calculator) and/or carries G25 coordinates (usable for nmonte_fit, nearest_populations and run_calculator), whether the coordinates are simulated (from an admixture calculator, less accurate than computed Geno25 coordinates), the testing company, and when it was saved. A "@" row (calculator_result) holds a raw-file calculator's percentages, not G25 coordinates. Free. How to read this: "simulated: true" coordinates come from a free calculator run and are a rough placement; computed coordinates from the raw file are the accurate kind. A RAW sample with has_g25 false has no coordinates yet: make_g25_from_raw makes them. No samples yet: add_raw_file gives the user a link to add their raw DNA file, or pass 25 pasted G25 coordinates to the tools.
Inputs: include_coords.
make_g25_from_rawMake G25 coordinates from a raw fileADestructiveInspect
Computes Global 25 (G25) coordinates from one of the user's own RAW samples (an uploaded raw DNA file, see list_my_samples or add_raw_file) and saves them as a new sample ":-geno25", which nmonte_fit, nearest_populations and run_calculator take as a target. Needs a Basic or Pro plan and one G25 set (annual plans include sets; more are bought with purchase kind "g25"); it spends no credits. Without confirm the call only checks the sample and the sets and says what will happen; tell the user it uses one G25 set and call again with confirm: true only after they agree. Takes a few seconds. How to read this: these are computed coordinates from the genotypes, the accurate kind, unlike the simulated coordinates a calculator run gives; a sample that already has them is not computed again.
Inputs: sample (required), confirm.
nearest_populationsNearest populations to a targetARead-onlyIdempotentInspect
The populations whose G25 averages lie nearest a target, by pool (modern, ancient or both), with distances ×100. Free. How to read this: this is directional, not a match. A nearest modern population at distance 3 says where the target sits among living groups; a nearest ancient population at 8 is a hint about which ancestral pole a model should cover, not a source to copy. Use nmonte_fit to test an actual model.
Inputs: target (required), k, pool, label, +1 more.
nmonte_fitRun an nMonte fit (costs credits)ADestructiveInspect
Fits a set of source populations to a target exactly, where the target is pasted G25 coordinates (as on the /g25 workbench's custom line) or a saved sample: the objective Genoplot's nMonte optimises on the /g25 workbench (non-negative proportions summing to 100% minimising the distance to the target in the 25 G25 dimensions), solved exactly rather than by Monte Carlo, over the same reference rows. Costs credits (get_credit_balance shows the price; students spend purchased credits). Sources are catalogue names or ids from search_populations ("Russia_Samara_EBA_Yamnaya", "Russia_MLBA_Sintashta|22"), single reference individuals ("Russia_MLBA_Sintashta|22:I0939"), or the user's own saved samples (":"), at most 16. The penalty switch favours sources close to the target, as on the workbench. Returns the distance, its band, each source's share and id, the sources that took no weight, the model's period, and warnings. How to read this: the distance is Euclidean in the 25 G25 dimensions, ×100 as Genoplot shows it, and Genoplot's bands are ≤2 good, ≤3 fair, ≤5 poor, above 5 bad. It is relative: compare models of the same target with the same coordinate type only; scaled and unscaled distances are not comparable. Lower is not automatically truer: adding sources always lowers the distance, so a low distance with many sources is not evidence of ancestry, and populations that resemble the real ancestors fit almost as well as the real ancestors. A model draws all its sources from one period (modern or ancient; the result's period and warnings say when it does not), and a proximal ancient model from one horizon. Shares under about 2% are usually noise. The result's id values are the catalogue ids the Genoplot workbench accepts.
Inputs: target (required), sources (required), rows, label, +3 more.
purchaseSubscribe or buy creditsADestructiveInspect
Prepares a purchase. kind "subscription" with a plan (basic_monthly, basic_annual, pro_monthly, pro_annual) mints a Chargebee checkout link the user opens in their browser to pay; nothing is charged until they complete it there, and purchase_status confirms it afterwards. An account that already has an active subscription is refused (has_subscription): plan changes are made in the app. kind "credits" buys credits in units of 500 with the card already on file for an active subscription (no subscription: offer a subscription first), charged immediately. kind "g25" buys G25 sets the same way (one set makes G25 coordinates from one raw file with make_g25_from_raw). Without confirm the call only returns the quote (what, how much, what happens); ask the user and call again with confirm: true only after they explicitly agree. Never say a purchase is done until purchase_status (subscription) or this call's result (credits, G25 sets) says so. How to read this: prices are in US dollars, and a sale running in the app applies here too (price_usd is what is charged, list_price_usd the regular price); a subscription refills credits every period and unlocks Admix Plus (qpAdm); purchased credits are added on top of the balance and, for students, are the credits that connected assistants can spend.
Inputs: kind (required), plan, confirm, quantity.
purchase_statusConfirm a subscription checkoutAIdempotentInspect
Confirms whether a Chargebee checkout prepared by purchase (kind subscription) was completed, and if so records the new plan on the Genoplot account, exactly as the app does after its checkout. Call it when the user says they have paid. How to read this: "not_complete" means the user has not finished paying yet (ask them to complete the checkout, or wait); "subscribed" means the plan is active and credits follow from Genoplot's billing webhook within a minute.
Inputs: hosted_page_id (required).
run_calculatorRun a G25 calculator (costs credits)ADestructiveInspect
Runs a G25 admixture calculator on one target, the same pipeline as the /calculators workbench with its default settings (200 batches, 200 cycles, penalty on, outliers filtered). The target is pasted G25 coordinates (an array of 25 numbers, or a line "Name,c1,…,c25" as the workbench's custom field takes it; nothing is saved) or a saved sample: one of the user's own with G25 coordinates (list_my_samples), one shared with them, or a reference individual. The calculator is an id or name from list_calculators. Takes up to about 25 seconds. Returns the component percentages, the fit distance, the nearest reference populations and which coordinate type was used. Charged by the pipeline at Genoplot's standard calculator rate (see get_credit_balance). How to read this: a calculator breaks the sample into the calculator's own components (its "K"), which are abstract clusters chosen by the calculator's author, not real populations; the percentages depend on the calculator and are comparable only with other results from the same calculator. The fit distance is the calculator's own scale, not the nMonte bands. Simulated coordinates (from a free calculator run) give rougher placements than computed Geno25 coordinates. The run is charged by Genoplot's pipeline at its standard calculator rate.
Inputs: target (required), calculator (required), label, scaled.
run_pcaRun a PCAARead-onlyIdempotentInspect
Runs the PCA the Genoplot /pca workbench draws, on G25 coordinates: the axes are fitted to the individuals of the chosen reference populations (their Average rows only projected, as the workbench does), and the user's samples or pasted coordinates are placed on them. Free. Choose populations by catalogue name or id (search_populations finds them; a single reference individual is "Group|v:Sample"), and/or curated regions, which add the region's modern populations, or its Bronze Age, Iron Age or distal source populations with region_set. At most 80 populations and 12 samples. Samples are inside the fit by default, as on the workbench; project_samples fits the axes on the populations alone and places the samples on them, which keeps one unusual sample from bending the axes. Returns each component's share of the variance, every population's centre and spread, each sample's scores and its five nearest population centres, and with individuals: true every reference individual's scores (sampled down past 1500). How to read this: a PCA's axes are fitted to the populations chosen, so the same sample sits elsewhere on a different selection, and positions are only comparable within one run. PC1 and PC2 carry the share of the variance explained_variance gives, and anything they do not carry is invisible on a 2D plot: two points close on PC1/PC2 can be far apart in the other dimensions (nearest_populations measures all 25). nearest is the distance ×100 to each population's centre over the returned components only. Being near a population on a PCA is not descent from it: an admixed sample sits between its sources, and a population that resembles the real ancestors sits where they would. Axis signs are arbitrary (a mirror image of the workbench's plot is the same plot). A population's spread is the standard deviation of its individuals on each axis; a sample beyond about two spreads from a centre is not typical of that population.
Inputs: scaled, regions, samples, components, +5 more.
run_qpadmSubmit a qpAdm model (costs credits)ADestructiveInspect
Submits a qpAdm model (ADMIXTOOLS 2 through Genoplot's formal-statistics service): a target, a list of source populations ("left") and a list of outgroups ("right"), on a reference dataset. The target is the user's own uploaded RAW sample (":" from list_my_samples; it must have a raw file, G25 coordinates alone are not enough) or a reference population label from the dataset (e.g. "Sardinian.HO"; set target_is_individual for a single reference individual). Populations use the dataset's labels with their suffix (.HO, .DG, .SG, .AG). The run takes one to several minutes and this call returns a job id at once: poll get_qpadm_result until status is terminal. Requires the Admix Plus entitlement (Basic or Pro plan), accepted terms, and a minimum balance (get_credit_balance); the run itself is charged by the service at Genoplot's standard rate, and a run refused for credits appears as an error in get_qpadm_result. At most 2 jobs run at a time. How to read this: the p-value is the probability of observing data like the user's if the model were true. It is not the probability that the model is true, and it is not a score of how good the model is. Above 0.05 the model is not rejected (the data are consistent with it); below 0.05 it is rejected (the sources cannot account for the target's relationships to the outgroups); a p-value near 1 often means the outgroups lack the power to distinguish anything, not that the truth was found. Not rejected is not the same as correct: several mutually exclusive models routinely pass for the same target and the statistics cannot choose between them. "feasible: false" (a negative coefficient) means the model is mathematically inadmissible even if its p-value is high. Standard errors are the honest width of each weight: 54% ± 10.5 means roughly 43 to 64%. Coverage matters: a consumer chip calls under a tenth of the reference SNPs, which widens the errors; a comfortable p-value on a low-coverage file is weaker evidence than the same p-value on a dense or imputed one, and with allsnps true p-values are anti-conservative when outgroups vary in coverage, so re-run a borderline model with allsnps false.
Inputs: left (required), right (required), target (required), allsnps, +2 more.
run_raw_calculatorRun an admixture calculator on a raw DNA file (costs credits)ADestructiveInspect
Runs a GEDmatch-style admixture calculator (Eurogenes, Dodecad, HarappaWorld, MDLP, Tolan and others) on one of the user's own raw DNA files (a RAW sample from list_my_samples, ":"), as the /calculators workbench does in raw-file mode, and returns the component percentages in the same call. Every calculator the app offers for raw files is here, e.g. Eurogenes K13, K15 and K36, Dodecad K12b and Globe13, HarappaWorld, MDLP World K22, Lukas M K47; list_calculators with kind "raw" lists them all with their components. A file that already ran the calculator returns that saved result at once and is not charged, as in the app; force_recalc: true runs it again anyway (the app's Force recalc switch), charged like any run, and the new result replaces the saved one. Otherwise the run usually takes 10 to 20 seconds and is charged by the run itself: about 20 to 30 credits, depending on how long the run takes (get_credit_balance shows the balance; a run needs at least 30 to start). After a fresh run of Eurogenes K13, K15 or K36 (or their Ancient versions), HarappaWorld (or Extended), MDLP World K22, Lukas M K47 or Michal World K25 the app also saves a simulated G25 sample ("-K13-sim" and so on), and so does this tool; nmonte_fit, nearest_populations and run_calculator take it. Files added with a personal encryption token, and files shared by someone else, run in the app only. How to read this: an admixture calculator splits the genotypes into its own components (its K), clusters its author built from reference samples, not real populations. The same file gives different percentages in different calculators, so compare results only within one calculator, and treat components under about 2% as noise. snps.share_pct is how much of the calculator's SNP set the file covers; the lower it is, the less stable the percentages. A simulated G25 sample made from these percentages is a rough placement, not coordinates computed from the genotypes.
Inputs: sample (required), calculator (required), force_recalc.
search_guidesSearch the Genoplot guidesARead-onlyIdempotentInspect
Searches Genoplot's 11 written guides (uploading a DNA file, admixture calculators, getting G25 coordinates, nMonte modelling, qpAdm and f-statistics, PCA, the health dashboard, family trees, credits and subscriptions, saving and sharing) and returns the best-matching sections with an excerpt and a link straight to the section. Free. How to read this: quote or paraphrase the excerpt and give the link; the guides are the authoritative description of how Genoplot works and what runs cost.
Inputs: query (required), limit.
search_populationsSearch reference populationsARead-onlyIdempotentInspect
Finds populations in Genoplot's G25 reference catalogue by words in the name, place or ethnicity, optionally only modern or ancient, or dated within a year range (negative = BCE; most entries are undated and match a range by their horizon). Plain words work: periods ("neolithic", "bronze age" find the N, EBA, MLBA in names), regions ("levant", "anatolian" find the countries filed under them) and the classic poles ("western hunter-gatherer", "steppe"). Every word must match; the best matches come first. Returns each population's catalogue id ("Group|v"), name, sample count, period, horizon (hunter-gatherer, neolithic, chalcolithic, bronze, iron, historical, medieval), origin and date. Use it to turn a people or culture the user names ("Sintashta", "Levant Bronze Age", "Sardinian") into source ids for nmonte_fit or populations for run_qpadm. How to read this: a hit is a catalogue entry, not a claim about the user; "simulated" populations stand in for unsampled ancient poles and belong to ancient models only.
Inputs: query (required), era, limit, to_year, +1 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Related MCP Connectors
Public read-only MCP for the Y-chromosome haplogroup tree & Y-SNP resolution. 父系单倍群树查询与Y-SNP解析。
Human Lineage MCP server — search and query the public genealogical graph of all humanity
Pedigree charts from GEDCOM
Ancestry allele frequencies, LoF constraint, variant lists, and coverage from gnomAD, with ClinVar.
Related MCP Servers
- AlicenseAqualityBmaintenanceUpload your raw DNA file from 23andMe, AncestryDNA, MyHeritage, FamilyTreeDNA, or any VCF -- and watch a team of AI agents fan out across 16+ public genomics databases, share discoveries with each other in real time, and produce a comprehensive health report. Everything runs on your machine. Nothing is uploaded anywhere.2416MIT
- FlicenseAqualityCmaintenanceEnables querying rare-variant, gene-based association results across ~1.2M individuals from 10 global biobanks, supporting phenome-wide scans, replication screens across ancestries, and candidate list evaluation for 44 harmonized traits.4-
- AlicenseBqualityBmaintenanceFamily history research toolkit with 79 archives and databases, name variants, historical calendars, kinship, evidence checks, GEDCOM auditing and archive-request drafts. An MIT-licensed Python MCP server that runs locally without API keys or network requests.31274 PyPI10MIT
- AlicenseNot gradedqualityCmaintenanceAccess to 1000 Genomes Project dataset, hosted online in Dnaerys variant store3Apache 2.0
Glama MCP Gateway
Add one secure layer between your agents and this server.