Early access — Compute markets intelligence

GreenCIO, native to your AI workflow.

Probability signals, GPU-hour price tape, decision briefs, and methodology — exposed through the Model Context Protocol so Claude and Cowork can answer compute-market questions inside the tools you already use.

Source detectedProbability movedDesk reviewedBrief ready
Get early access
See live signals

Why MCP-native

The terminal is moving inside your AI assistant.

Analysts and allocators don't alt-tab between Bloomberg, Slack, and a half-dozen web UIs anymore. They sit inside Claude or Cowork and pull data in as they think. GreenCIO ships an MCP server so compute markets data flows into that workflow the same way every other data source already does — no tab-switching, no copy-paste.

For analysts and capital allocators

Ask in plain English. Defend the answer with source trails.

You

What is the H100 spot price right now?

→ get_compute_index_snapshot

GreenCIO MCP

{
  "headline": {
    "indexName": "GCI-H100-USE-SPOT",
    "value": "<current_spot>",
    "asOf": "<generated_at>",
    "provisional": false
  },
  "methodologyStatus": "persisted",
  "methodologyVersion": "gci-v0.1.1"
}

You

Find Meta datacenter signals from this month.

→ search_intelligence

GreenCIO MCP

{
  "signals": [
    { "title": "Meta Louisiana datacenter financing closes at $29B", "impactLevel": "high" },
    { "title": "Meta files PUCO interconnection request in Ohio", "impactLevel": "medium" }
  ],
  "totalMatched": "<count>",
  "dataMode": "gcs"
}

You

Give me my morning brief.

→ morning_brief

GreenCIO MCP

# Morning brief

## Pulse (last 24h, impact >= medium)
- Meta Louisiana datacenter financing closes at $29B
- ERCOT interconnection queue cleared 4 new sites this week

## Compute Index
- Headline GCI-H100-USE-SPOT: <current_spot>/hr (settled, v0.1.1)
- USE vs USW spread: <basis_spread> (<spread_pct> of base)

> Provenance: feed=gcs, gci=persisted

Illustrative response shapes. Placeholder fields represent live values returned by the connected host; this page does not publish current prices. Every response carries data-mode and methodology-status flags so the chat agent can disclose provenance.

For engineering buyers and AI infra teams

Stdio MCP server. 11 typed tools. Deterministic output.

The 11 tools

  • get_intelligence_pulseWhat is happening in AI infra today.
  • search_intelligenceFind signals about a company, location, or technology.
  • summarize_signal_clusterSummarize what has happened with X over a period.
  • ingest_signal_from_urlAdd a news article to the feed.
  • get_compute_index_snapshotCurrent GCI spot settlement for a SKU and region.
  • get_settlement_historyTime-series of settlements for one index.
  • get_forward_curveForward-curve status for a SKU and region; warnings until points are exposed.
  • get_basis_spreadCross-index basis spread between regions or SKUs.
  • get_methodologyActive GCI methodology and rulebook hash.
  • explain_price_moveExplain a recent price move with related signals.
  • morning_briefDaily digest combining intelligence and compute.

Architecture

  • Stdio transport. Runs as a local subprocess; no inbound network surface.
  • Authenticated Data API. GCI tools require a tier-pro key; public feed tools work without one.
  • Structured errors. Every failure returns a machine-readable code: UPSTREAM_AUTH, SYNTHETIC_BLOCKED, LOCAL_RATE_LIMIT, and others.
  • Provenance built in. Feed responses expose dataMode; GCI responses expose methodologyStatus and methodologyVersion.
  • No invented data. Tools return structured errors or warnings instead of fabricated forward curves or settlements.
  • Rate-limited write surface. The single write tool (URL ingestion) is gated by a server secret and a local per-hour limit.

Claude Desktop / Cowork config

{
  "mcpServers": {
    "greencio": {
      "command": "node",
      "args": ["/absolute/path/to/greencio-mcp-server/dist/index.js"],
      "env": {
        "GREENCIO_BASE_URL": "https://your-greencio-host",
        "GREENCIO_DATA_API_KEY": "gc_pro_..."
      }
    }
  }
}

The pro-tier Data API key is minted by GreenCIO as part of early-access onboarding.

What you don't get from a UI alone

Defensibility for the answer the chat agent gives.

Source trails

Every signal carries source URL, source name, and discovery timestamp. The chat agent can quote evidence back into a memo without losing the citation.

Methodology versioning

Every GCI settlement reports the methodology version and rulebook hash. The get_methodology tool returns methodology metadata and the active rulebook hash on demand.

No silent demo data

When upstream returns synthetic data, the tool flags it as methodologyStatus="synthetic" — or blocks it entirely when STRICT_NO_SYNTHETIC is set.

Early access

What to expect right now.

When is it live?

The stdio MCP server is built and tested. Early-access onboarding is rolling out to design partners now; new keys are minted manually during this phase so we can match seat usage to your workflow.

What do I need?

A tier-pro GreenCIO Data API key and a recent Claude Desktop or Cowork install. We mint the key as part of the early-access conversation.

Which AI tools does this work with?

The server uses standard stdio MCP and is built for Claude Desktop and Cowork first. Other MCP-compatible clients may work, but Claude remains the release gate for tool-selection accuracy.

Is data sent anywhere new?

The MCP server runs as a local subprocess. It calls the GreenCIO Data API the same way the web console does, using your bearer token. No new hosting required on your side.

Compute markets data, where you already work.

Join the early-access cohort. We'll set up your key and the Cowork install in one short conversation.

Get early accessSee live signals →