MCP · Research access
Ask this dataset a research question, not a query.
Eight tools, each answering a question a researcher actually asks. Every answer arrives with the corpus it was drawn from, how much of that corpus carries each signal it used, and a list of what was not measured — so you can tell a quiet product from an unmeasured one.
Connecting
One URL and a key.
- Create a key on your account page. It is shown once. Keys require Pro.
- Add the server to your MCP client. In Claude Code:
claude mcp add --transport http magellantools https://magellan.tools/mcp \
--header "Authorization: Bearer apt_your_key_here"Any client that speaks Streamable HTTP works the same way: the endpoint is https://magellan.tools/mcp and the key travels as Authorization: Bearer …. Sessions are not accepted — a browser cannot reach this endpoint.
The eight capabilities
describe_datasetWhat this dataset can answer — coverage per signal, which ones are too thin to build on, and how the default ordering is computed. Start here.survey_marketThe shape of one category: size, distribution of momentum and authority, who leads on each axis, and how concentrated assistant referral is.find_productsState a thesis — market, momentum, stage, assistant demand, search posture — and get the products that fit, each with the measured values that qualified it.trace_trajectoryA year of daily link history for up to five products: the weekly series, 30/90/180/365-day changes, the steepest month, and whether growth is still accelerating.analyze_ai_demandWhich AI assistants send people where — by assistant, by product, or by category. No general web dataset answers this.analyze_search_demandA product's queries and brand posture, or a keyword's competitive set: every product measured against that query, ranked by clicks.brief_productEverything measured about one product, its standing against its category, its grounded answers — and, explicitly, everything that was not measured.compare_productsTwo to five products, compared only on dimensions where at least two carry a measurement. The rest are named as not comparable rather than shown as blanks.What it is good at
24,865 independent products, measured to different depths.
This corpus is strong on the link graph and on AI assistant demand, and weak on traffic analytics. Ask it about momentum, discovery and which assistants serve a market. Do not ask it for traffic volume — the answer would be blank for most of the corpus, and a blank is never filled in with an estimate.
Too thin to build a conclusion on
Reported when a product happens to carry one. Never aggregated, never ranked on, and never offered as a filter.
Questions it answers well
find_productsanalyze_search_demandanalyze_ai_demandtrace_trajectoryWhat every answer carries
findingThe answer in prose, numbers formatted for a reader.evidenceThe measured values behind it.corpusHow many products were searched, how many came back, and which corpus.coverageFor each signal used: how many products carry it, out of how many.not_measuredWhat was looked for and not found, by name. Absent, never zero.snapshotWhich snapshot answered, and the measurement window.linksPages on this site where the answer can be checked.Rate limits, the record budget and the audit trail are the same ones the paid API runs under; the methodology page explains how each signal is measured.
