FactIQ

Use FactIQ when you want an agent citing official statistics instead of recalling them. It replaces a stack of separate government APIs with one connection, and its earnings tools return structured claims rather than raw transcripts. At 16 GitHub stars it is early, so treat it as promising rather than proven.

FactIQ is an MCP server from Defog that gives AI agents read-only access to economic and financial data from 37 official sources. It connects over HTTP to api.factiq.com/mcp with one OAuth sign-in, and exposes discovery and fetch tools including run_sql, get_series and search_earnings_transcripts to Claude Code and Codex.

Maker: Defog · Protocol: MCP · Auth: oauth

Compatible agents: Claude Code, Codex CLI, Claude, ChatGPT, any MCP client

Required runtime: Claude Code or Codex CLI, A free FactIQ account

About FactIQ

Defog built FactIQ to turn a coding agent into a finance and economy analyst. Rather than shipping a chat product, it exposes a warehouse of official statistics through named tools the agent calls directly, spanning government releases, market data, earnings calls and satellite signals. The plugin repository is MIT licensed and was last updated in August 2026. It connects over MCP with HTTP transport to a single endpoint, api.factiq.com/mcp, and authenticates through one OAuth sign-in rather than an API key pasted into a config file. The agent works in two phases: discovery tools locate the right dataset and series, then fetch tools pull the numbers, including read-only SQL against schemas such as the Bureau of Labor Statistics, Census, BEA and SEC EDGAR. Some schemas are deliberately gated, so a query against the transcripts schema returns 403 rather than raw text. The natural user is someone already working in Claude Code or Codex who wants sourced numbers instead of a model's recollection of them. It suits macro research, trade flows by HS commodity, earnings-call analysis and satellite-derived signals such as nighttime lights, shipping activity and reservoir levels. It is a poor fit if you want a hosted dashboard, because there is no UI: the output is whatever the agent renders, from a terminal preview to a local HTML chart. Access is tiered: the entry level costs nothing, and paid plans buy history depth and breadth rather than extra software. Coverage runs deepest across the United States and Asia, and company-level data thins out considerably below the largest listed names. The repository star count is a fair signal of how early this project is, despite the size of the warehouse sitting behind it.

Key Features

• Two-phase tool design: Splits discovery from retrieval so the agent locates a dataset and series first, then fetches, which cuts the wrong-series errors that plague single-endpoint data APIs.
• Read-only SQL over official schemas: Exposes named schemas including BLS, Census, BEA, EIA and SEC EDGAR, so an agent can compute a derived metric rather than fetch a single prepared number.
• Structured earnings claims: Decomposes calls into a claim graph and separates management statements from analyst hypotheses, so an agent cannot attribute an analyst's framing to the company.
• Satellite-derived series: Carries nighttime lights, shipping and port activity, reservoir levels and fire detections for cases where official statistics lag or cannot be trusted.
• One OAuth connection: Authorises through a single browser sign-in from the client rather than distributing an API key into a local config file.
• Gated schemas fail loudly: Admin-only schemas are absent from your catalog and return 403 on access, so an agent gets an error instead of a silently partial answer.

Use Cases

• Pin an earnings claim to the exact quarter: Checks live transcript coverage first, pins the fiscal quarter, then pulls bounded management-claim and Q&A-pressure rows for that same call instead of guessing from memory.
• Trace a trade flow by commodity: Queries monthly imports and exports by HS commodity and partner country from US Census and China customs data, without mixing incompatible preliminary and per-HS datasets.
• Read an economy from orbit: Pulls satellite-derived series such as nighttime lights, shipping and port activity, reservoir levels and crop fires when official statistics lag or are unreliable.
• Build a sourced chart in the terminal: Runs the analysis and returns a terminal preview, report JSON or a bespoke local HTML visualisation, with every figure traceable to its source series.

Install

/plugin marketplace add defog-ai/factiq-plugin

Requirements

• Claude Code (or Codex CLI) with plugin support
• Install the plugin: /plugin install factiq@factiq, then run /reload-plugins
• A free FactIQ account at factiq.com (email, Google or passkey)
• Authorise the MCP server: run /mcp, select factiq, choose Authenticate

Actions

search_datasets

Finds which datasets in the warehouse cover a topic before any query is written.

/factiq:ask What US datasets cover job openings?

• query (string) — required: Free-text description of the topic to locate.

get_data_catalog

Returns the live, authoritative catalog of datasets available to your account, including per-dataset descriptions.

/factiq:ask Show me the full data catalog

run_sql

Executes read-only SQL against the warehouse schemas such as bls, census, bea, eia and sec.

-- read-only; the gated transcripts schema is never queryable this way
SELECT * FROM bls WHERE series_id = 'LNS14000000' ORDER BY period DESC LIMIT 12;

• sql (string) — required: A read-only SQL statement against a permitted schema.

get_series

Fetches the observations for a specific time series once its identifier is known.

/factiq:ask Get the last 5 years of US unemployment series LNS14000000

• series (string) — required: Series identifier, for example LNS14000000 for US unemployment.

search_earnings_transcripts

Retrieves structured claims and Q&A pressure points from earnings calls; it never returns a raw transcript.

/factiq:ask What did Micron management say on its latest call?

• search_target (string) — required: One of claims, pressure_points, disclosure_profile or coverage.
• query (string): Free-text theme to search within the selected target.
• company_filter (string): Restrict to a company; disclosure_profile takes the first ticker only.
• quarter_filter (string): Pin the fiscal quarter; ignored by disclosure_profile and coverage.
• claim_family (string): Group of related claims to filter on.
• limit (integer): Maximum rows returned.

search_media_appearances

Searches precomputed claims and passage cards from executive podcasts, TV interviews and conference appearances.

/factiq:ask How has this CEO talked about pricing in interviews this year?

• search_target (string) — required: One of search, claims, passages, pressure_points, appearances or coverage.
• query (string): Free-text theme; retrieval is strict lexical, then loose, then trigram.

How to Invoke

Exposed as MCP tools over HTTP that the agent calls by name once the plugin is installed and the server authorised. Discovery tools are search_datasets, describe_dataset, search_series and get_data_catalog; fetch tools are run_sql (read-only), get_series, get_market_data, search_earnings_transcripts, search_media_appearances and search_news. A /factiq:ask slash command wraps the whole workflow, and any economic question in a normal session auto-invokes the bundled skill.

Pricing

Free tier covers basic global macro and company data. Research is $250 per month and adds five years of history for S&P 500 companies. Institutional is $1,500 per month with 10 or more years of history and the broadest coverage. The plugin itself is MIT licensed and free; the subscription buys data depth, not software.

Strengths

• Covers 37 official sources behind one connection, replacing a pile of separate government APIs each with its own auth and format.
• Separates management speech from analyst framing in earnings data, a distinction most transcript tools collapse and which changes what you can honestly quote.
• MIT licensed plugin with a free account tier, so the integration can be evaluated end to end before any spend.

Weaknesses

• SEC coverage stops at roughly 950 US-listed companies at ten billion dollars or more in market cap, so small and mid-cap research falls outside it.
• 16 GitHub stars as of August 2026 makes this a very young dependency for anything load-bearing.
• No hosted interface at all: without Claude Code or a Codex CLI in front of it there is nothing to look at.

View the official FactIQ skill page