Connect Skyread to your AI assistant
Skyread: commercially licensed, analysis-ready public data, mostly US (World Bank for other countries): consumers and local markets (every ZIP code, city, county, metro area and state), retail and e-commerce, restaurants and quick-service, wine, beer and spirits, grocery and CPG, food commodities, housing, freight and supply chains, jobs, inflation, rates and public companies. Every value is sourced, dated and credited.
Get connected
Check you have what's needed, then follow the steps for your assistant.
What you need
- A Skyread account (start a trial). Assistants that support sign-in (Claude, ChatGPT) ask you to approve access once; others use your API key.
- The server address:
https://skyread.fundas.ai/mcp(MCP, Streamable HTTP). - Every tool only reads data, except creating and deleting your own alerts.
Connect your AI assistant
Skyread speaks MCP, the open standard for assistant tools. Pick yours; menu names can differ slightly between versions.
- Claude (web or desktop) → Settings → Connectors → Add custom connector. On Team and Enterprise plans an owner adds it for the organization first (Admin settings → Connectors); members then connect.
- Name: Skyread. URL:
https://skyread.fundas.ai/mcp - Click Connect; on the Skyread page that opens, sign in with the code emailed to your work address (or one of your API keys) and approve.
- In a chat, enable Skyread and ask: "Are restaurant margins being squeezed?"
- ChatGPT → Settings → Apps & Connectors → Advanced → turn on Developer mode. Custom connectors need a paid ChatGPT plan that allows them (not Free); on Business and Enterprise a workspace admin may need to allow them first.
- Create a connector: name Skyread, MCP server URL
https://skyread.fundas.ai/mcp, authentication OAuth. - Sign in on the Skyread page with the code emailed to your work address (or one of your API keys), then pick Skyread in a new chat.
Add to ~/.cursor/mcp.json (Windsurf: ~/.codeium/windsurf/mcp_config.json):
{ "mcpServers": { "skyread": {
"url": "https://skyread.fundas.ai/mcp",
"headers": { "Authorization": "Bearer YOUR_SKYREAD_KEY" } } } }Add to .vscode/mcp.json, then use Copilot Chat in Agent mode:
{ "servers": { "skyread": {
"type": "http", "url": "https://skyread.fundas.ai/mcp",
"headers": { "Authorization": "Bearer YOUR_SKYREAD_KEY" } } } }Gemini CLI: add to ~/.gemini/settings.json. Google Antigravity accepts the same server in its MCP settings.
{ "mcpServers": { "skyread": {
"httpUrl": "https://skyread.fundas.ai/mcp",
"headers": { "Authorization": "Bearer YOUR_SKYREAD_KEY" } } } }claude mcp add --transport http skyread https://skyread.fundas.ai/mcp \ --header "Authorization: Bearer YOUR_SKYREAD_KEY"
Building on the OpenAI API? Give a model Skyread as a remote MCP tool in the Responses API; it calls Skyread's tools while answering. Uses your OpenAI key for the model and your Skyread key for the data (keep both on your server, never in a browser). No ChatGPT plan is involved.
from openai import OpenAI
client = OpenAI() # reads OPENAI_API_KEY
response = client.responses.create(
model="YOUR_OPENAI_MODEL", # any model that supports remote MCP tools
tools=[{
"type": "mcp",
"server_label": "skyread",
"server_url": "https://skyread.fundas.ai/mcp",
"headers": {"Authorization": "Bearer YOUR_SKYREAD_KEY"},
"require_approval": "never", # Skyread's tools only read data, except creating/deleting your own alerts
}],
input="Are restaurant margins being squeezed? Cite the sources.",
)
print(response.output_text)
Each answer carries the publishers' credit lines; show them with any figure you publish.
Any tool can call the REST API (docs: https://skyread.fundas.ai/docs):
curl -H "Authorization: Bearer YOUR_SKYREAD_KEY" \ https://skyread.fundas.ai/api/v1/briefings/restaurants
What to ask, and what you get back
Your assistant picks the right tool for each question. This is the map it follows, and the rules every answer keeps.
Which tool answers which question
Assistants receive this guide automatically when they connect (also available as the gateway://guide resource and the "How to use Skyread" prompt).
| You want to know | For example | Tool |
|---|---|---|
| An industry or the economy: how is it doing, what is changing | "Are restaurant margins being squeezed? How are US consumers doing?" | get_briefing(topic); list_industries first if unsure which topic |
| A full industry report: size, five-year performance, costs, companies, states, outlook | "Write me an industry report on quick-service restaurants. What's the outlook for homebuilders in Texas?" | get_industry_report(industry, area optional): qsr, full_service, grocery, homebuilders, building_products, trucking, data_centers; outlook is scenarios, never a forecast |
| What bears on an industry: costs, demand, upstream forces, companies, where it is concentrated | "What should a trucking fleet watch? Which costs matter most for quick-service restaurants?" | trace_impacts(industry); explore_graph(node) for any industry, company, place, concept or series and its links |
| Online retail and e-commerce: online share, sales by store type, shipping costs | "What share of retail is online? Are clothing store sales growing? Are parcel rates rising?" | get_briefing(topic='ecommerce'); single measures with get_metric_series (census.ecommerce_share, census.clothing_store_sales, bls.ppi.couriers_express) |
| A place: who lives there, what they earn, how big the market is, what gas costs there | "What does the market in ZIP 10021 look like? How affluent is the Denver metro? What does gas cost in Phoenix?" | get_local_profile(area): ZIP code, city, county, metro area or state; its gas_price uses the closest area that EIA (weekly) or BLS (monthly) publishes and says which |
| The best or worst places by a measure | "Which ZIP codes around Chicago have the most $200k+ households?" | rank_areas(metric, level, within) |
| TV markets (DMAs) | "How big is the Atlanta TV market?" | use the metro area (OMB Core-Based Statistical Area) with get_local_profile; DMA definitions are proprietary |
| One number or its trend | "Latest CPI for eggs? Diesel price? How many jobs were added?" | get_metric_series(metric_id); find ids with explore_ontology_nodes, place ids with find_places |
| Today's dollars, or what was known on a past date | "Restaurant sales in 2019, in today's dollars?" | get_metric_series(metric_id, adjust='real') or get_metric_series(metric_id, as_of=...) |
| Prices in a detailed category | "How fast are wine prices at bars rising?" | get_granular_cpi(category) |
| Countries | "Compare GDP growth in the US, China and India" | query_global_macroeconomics(country_code, indicator) |
| Housing | "Is housing getting less affordable?" | analyze_housing_market() |
| Supply chains and freight | "Is anything disrupting US supply chains now?" | analyze_supply_chain_shocks() |
| Public companies | "Compare McDonald's and Yum margins" | get_company_fundamentals(tickers) |
| Cruise ship sanitation | "Which cruise lines failed inspections?" | fetch_vessel_health_profiles(cruise_line) |
| Franchise and small-business loan risk (SBA 7(a)) | "Which franchise brands have the most SBA loan defaults? How does Subway compare with other sandwich shops?" | get_franchise_loan_outcomes(franchise or naics, state optional): charge-off rates, mature loan years only |
| Anything else in the data | "Monthly beer imports by port" | list_datasets(), then run_sql (read-only) |
| Keep watching something | "Tell me when diesel is up 10% on a year ago" | create_alert, list_alerts, get_alert_events; delete_alert(alert_id) to stop one (confirm with the user first) |
Rules every answer follows
- Quote the unit, the period each value describes and the publisher; values are as published and may be revised.
- Include attribution.credit_line whenever you quote or summarize a value: the publishers' terms require it.
- Survey values (census.*) carry margins of error: quote them for small areas; never rank or compare figures flagged unreliable. 'Higher' or 'lower' in a local profile means the difference passed Census's 90% significance test.
- Reference-only data points (reference_only=true) are not stored because their terms forbid it: they return no values and a how_to_get_it block. Tell the user where to get the data and offer the alternative it names; never fill the gap from memory.
- Every payload has plain_summary, how_to_read and caveats: use them to explain results to non-experts.
- Skyread is descriptive data and analysis, not investment, financial, tax or legal advice, and not a forecast.
All tools
REST API reference: https://skyread.fundas.ai/docs. How accurate the answers are: public accuracy test.
What Skyread sees
Your data
Skyread stores your company name, email, Stripe billing references, hashed API keys and sign-in tokens, daily usage counts, and the alerts you create (with the email or webhook address you give them), plus the encrypted push address of any device you turn alert notifications on for. Our host keeps standard server logs (IP address, time, address requested) for up to 30 days. Skyread never receives your conversations, chat history, memory or files, only the tool requests your assistant sends, and it does not keep those. See the privacy policy. Questions: skyread@fundas.ai.