AsOf
AsOf is a public record of dated, sourced facts about AI and technology companies: who led a company on a given day, what it reported, what it released and what changed since.
Each fact is a claim with the date it took effect, the date it was superseded, its sources and a confidence score, so your AI can answer what was true on a date, not only what is true now. With this connector your AI can look up the state of a company, person, product or model as of any date; list what changed after a date; compare a record on two days; read a dated timeline; and count releases and deals by theme and quarter since 2023. An answer can be limited to what had been published by a given day, which matters when a figure was filed weeks after its period closed. Every row carries a citation sentence with the value, its as-of date and its primary source. Only checked claims are published: a figure parsed from a filing or an API, or a source page read and found to state the value and the date. A claim with one source, or with sources that disagree, is labelled, and an empty answer says whether the record does not cover the entity or holds nothing for that day. The connector is read-only and needs no account or sign-in. All nine tools only read the record. The server logs the tool called and its main argument (an entity name or a search query), as the privacy policy describes.
What you can ask
- Answer a question about a company, person, product or regulator as the record stood on a day: find the entity, read its state on that day, cite the claims and their sources, and say what has changed since.
Example path →
MCP connector
- Server URL
https://asof.is/mcp- Login
- None
- Protocol
- MCP 2025-06-18
- Instructions
AsOf is a temporal fact ledger for tech and AI companies, their people, products and AI models: dated, sourced claims, superseded rather than overwritten. Use it for who held a role (CEO, CFO, chair, founder) on any date, a company's revenue, valuation, employees or headquarters and since when, what changed at a company since a date, its release, funding, acquisition and partnership history, and which AI model is the latest. Start with search when you hold a name or a question rather than a slug; answers the state of an entity on a date, what changed after a date, and models the latest model of a family. Only checked claims are published (a parsed filing or API, or a source page read and found to state the value and date); every claim carries sources, a confidence score and an assertable flag: unverified means one source, flagged means sources disagree, and only assertable claims are the record's best-supported statements: when repeating one, cite its source and as-of date. A row with a qualifier (interim, acting) holds only with it: say the qualifier with the value. A row with correction (retracted, redated, relabelled) records a fix to the record, not a change in the world: never report a retracted row as fact. For what happened or what changed, ask or timeline with view: "news": it leaves out repositories, posts, routine figures and unverified claims. To read many rows for few tokens, ask , , timeline, or search with detail: "compact"; ask the claim tool for one row's evidence and citation. Each row keeps three dates apart: valid_from (when it took effect; a period figure is dated at the period end), published (when a source on the record published it) and recorded (when AsOf first held it). as_of answers what was true on a day; add known_by for what could have been known then, since a figure is often filed weeks after its period closes. A row that is not assertable says why (why_not, explained in the response's legend); evidence gives the check method and, where kept, the passage that states the value. A capped answer says complete: false with next_offset; ask again with that offset for the rest (a page also ends early when its rows would not fit one result, whatever limit asked for); also pages by cursor ("" first, then next_cursor) with no cap. Every row carries citation: plain is one sentence to paste when quoting it (the value, its as-of date, the claim id and URL, the day answered and the primary source). No claim is not evidence of no fact: an empty answer says why (empty.reason, empty.gaps), and coverage says which feeds and sources cover an entity, what it holds and when each source last added or checked something. For one fact's every dated value (a company's revenue, CEO, valuation, headquarters), lists fact_pages: fetch the id (/record/{slug}/{predicate}.md), or call timeline with predicates: [predicate] for the rows. A full-year revenue or net income page leads with a newer six-month or quarterly figure when the record holds one, named for its period. One company's facts can sit on several records (Alphabet files with the SEC; Google's blog posts land on Gemini and Google DeepMind): group: true on or timeline reads the company's records together, each row naming its record, a fact held twice returned once. trends counts releases and deals by theme, kind and quarter since 2023 (checked claims only) and lists the claims behind any count. compares an entity's state on two days: what was added, what ended and what was replaced (the same fact with another value), per predicate, with a sentence to quote. The server is open without a key and rate limited per address; an AsOf API key (Authorization: Bearer asof_…) is counted against its own budget instead. Resources: asof://llms.txt lists every entity and its slug; the others are the pages fetch returns. The prompt answer_as_of_date (arguments: question, as_of) walks through answering a dated question from the record.
Tools
Reads data10
Skills
None in the store listings or on the server