@compute-finance/mcp

@compute-finance/mcp

Live AI compute pricing oracle — real-time LLM model prices across providers (Anthropic, OpenAI, Google, xAI) via the Compute Finance Oracle.

A stdio MCP server. Works in any MCP client. Includes optional Claude Code skills for session cost analysis.

Quick start

Claude Code (recommended)

npx @compute-finance/mcp setup

This single command:

• Registers the MCP server at user scope (claude mcp add)
• Installs Claude Code skills (/cf-session-management, /cf-session-consumption, /cf-active-sessions)
• Installs the cost hook — a UserPromptSubmit hook that injects session cost into Claude's context so every response can show how much you've spent

Restart Claude Code after setup.

Or register manually without skills/hook:

claude mcp add --scope user compute-finance -- npx @compute-finance/mcp

Cursor / VS Code / Any MCP client

Add to your MCP config (.cursor/mcp.json, VS Code settings, etc.):

{
  "mcpServers": {
    "compute-finance": {
      "command": "npx",
      "args": ["@compute-finance/mcp"]
    }
  }
}

From source

git clone https://github.com/compute-finance/mcp.git
cd mcp
npm install && npm run build
npx . setup

Tools

17 tools across five layers — no API key required. All tools are read-only.

Data (live oracle)

Tool	Description
data_get_basket	All models with provider, family (e.g. openai.gpt, anthropic.claude), base_* and billed_* USD prices per million tokens, and per-component cache pricing (read, write-5m, write-1h) with provider attribution
data_get_price	Price for a single model (e.g. claude-opus-4.7) — basket members and catalog-only entries on identical terms
data_get_scu	Current Standard Compute Unit — value plus a methodology-versioned breakdown listing every family representative
data_get_breakdown	Per-family blended-cost breakdown alone — methodology-versioned discriminated union with one entry per family representative
data_get_cpi	Full Compute Price Index — basket with scuUsd, version, raw/marked-up prices
data_get_reconstitutions	Historical basket changes — model swaps, SCU before/after
data_get_methodology	Methodology changelog — every version with its formula summary and spec link, plus the version in force
data_get_history	SCU index time series over a date range — per-revision, daily, or weekly granularity; daily/weekly buckets carry the last revision's value forward across empty buckets
data_get_model_price_history	Per-model input/output USD price time series for a model that has appeared in the SCU basket — same granularity semantics as data_get_history, with catchup gaps surfaced in unavailableRevisions

Cache pricing comes from the Compute Finance Oracle. Session and consumption reports show effective (cache-aware) cost when the oracle has published the relevant cache components; otherwise they show nominal cost (input rate applied to every input variant) and label effective as unavailable for that model.

Every price is reported on two bases: base_* is the provider list price, identical for every model the oracle tracks, and billed_* is what compute.finance charges — base × (1 + routing_fee_rate). Compare models on base_*, budget on billed_*. The rate ships once per response and billed_* is null when the oracle does not publish it. Session and consumption reports are on the base basis throughout.

Compute

Tool	Description
compute_estimate	Nominal USD cost for a model given input/output token counts — base_usd_cost, routing_fee_usd, billed_usd_cost
compute_compare	Rank all basket models by cost for a workload, grouped by family — the same three cost figures per row

Render (Claude Code skills)

Tool	Description
render_session_report	Pre-formatted session cost report — used by /cf-session-management
render_consumption_report	Pre-formatted per-inference breakdown — used by /cf-session-consumption
render_active_sessions	Overview of recent sessions across projects — used by /cf-active-sessions

Reports surface three orthogonal counts: prompts (what you typed), inferences (assistant replies — tool-loop sessions produce several per prompt), and tool calls (tool_use blocks). The triplet is identical across all three reports for the same session.

Analysis

Tool	Description
analyze_session	Raw JSON session analysis (for custom UI, not skills)
analyze_inferences	Raw JSON per-inference breakdown (for custom UI, not skills)

History

Tool	Description
telemetry_get_history	Aggregate stats across logged sessions — cumulative cost, per-profile medians, insights

Cost hook

The setup command installs a UserPromptSubmit hook into ~/.claude/settings.json. Every time you send a message, the hook reads the current session transcript, prices it against the live oracle, and injects a cost summary into Claude's context via additionalContext. Claude then appends a 💰 Compute.Finance · … line at the end of its response.

Guards — the hook fires only when all three conditions are met:

• Session cost exceeds $1
• Session has at least 5 user prompts
• At least 10 minutes since the last fire (per session)

On any failure (oracle down, transcript missing, parse error) the hook exits silently — it never blocks your prompt.

Manual installation

If setup can't write to settings.json, add the hook manually:

{
  "hooks": {
    "UserPromptSubmit": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "command",
            "command": "npx @compute-finance/mcp hook-prompt"
          }
        ]
      }
    ]
  }
}

Uninstall

Remove the UserPromptSubmit entry from ~/.claude/settings.json.

Privacy

All data stays on your machine. The only network calls are unauthenticated GETs to api.compute.finance/v1/oracle/*. Session logs (~/.compute-finance/sessions.jsonl, ~/.compute-finance/inferences.jsonl) are never uploaded.

Links

• Compute Finance
• Oracle API
• OpenAPI spec
• npm package