For AI agents

MCP server and open API: barometer data as first-class tools for AI assistants

What this is

AI Investor Barometer tracks, daily, how five LLMs (GPT, Claude, Gemini, DeepSeek, Grok) value the same 24 Helsinki- and US-listed companies. The models produce only assumptions — a deterministic DCF engine computes the estimates. Through the MCP server your AI assistant reads these figures straight from the database: the numbers never come from a language model at query time — they come from the measurement pipeline, with as-of dates and engine versions attached.

What makes the data unique: this is not a stock-data API but a meta-layer about AI models themselves — what the models believe, how much they disagree, how well calibrated they are, and how their view differs from human analysts. A research observatory, not investment advice.

Setup

Remote server (Streamable HTTP), no API key, nothing to install. Endpoint:

https://ai-investor-barometer-production.up.railway.app/mcp

Claude Code

claude mcp add --transport http aiinvestorbarometer https://ai-investor-barometer-production.up.railway.app/mcp

Other MCP clients (JSON config)

{
  "mcpServers": {
    "aiinvestorbarometer": {
      "type": "http",
      "url": "https://ai-investor-barometer-production.up.railway.app/mcp"
    }
  }
}

Claude.ai: Settings → Connectors → "Add custom connector" and paste the URL above.

Tools

get_barometer_today

Latest snapshot: every covered company with the model-median estimate, valuation gap vs spot, model disagreement and the analyst comparison point.

get_company

Deep dive on one ticker: per-model estimates with the assumptions behind them (revenue CAGR, EBIT margin, WACC), reverse-DCF implied assumptions, disagreement rank.

get_company_history

Daily time series (up to 90 days): spot, panel median, bear/bull extremes, per-model estimates and assumption changes — with series-break metadata.

get_model_profile

Behavioral profile of GPT, Claude, Gemini, DeepSeek or Grok: bias, calibration, estimate volatility, validity rate, realized accuracy, current extremes.

get_disagreement_indices

Market-level AI indices: ACDI (median AI valuation gap), ADI (cross-model disagreement), ANM (sentiment shift) as daily series plus today's drivers.

get_ai_vs_analyst

The AI panel benchmarked against human analyst consensus on realized prices: systematic bias, optimal blend weight, up/down-market split.

get_signals

Weekly AI Signals reports and Weekend Read essays — list with TLDRs, or the full markdown body by id.

explain_methodology

How to read every number: vocabulary, engine-version series breaks, coverage, interpretation rules. Agents are instructed to call this before interpreting data.

Try prompts like

“What do the five AI models think about NOKIA right now, and how has their view changed over the last month?”
“Which AI model has been the most pessimistic on US mega caps, and is it better calibrated than the others?”
“How much do the AI models disagree with each other today compared to the current engine regime's history?”
“Summarize this week's AI Signals report and explain what drove the change in the disagreement index.”

Design principles

Every tool response carries the as-of date, engine version and a research disclaimer — an agent cannot present the data as advice or as fresher than it is.
Series breaks (engine versions, model changes) travel as machine-readable metadata in the history tools.
Vocabulary is descriptive: “model estimate” and “valuation gap” — never price targets or recommendations.
Read-only; the same public data as the website. Rate limit 60 requests/min/IP.
Usage is logged as tool name, duration and a salted IP hash — raw IPs are never stored.

Plain REST alternative

The same data is available from the public read-only REST API without MCP: OpenAPI docs at https://ai-investor-barometer-production.up.railway.app/docs. Site map for language models: /llms.txt.