TANDIU AI infrastructure data

Measured conclusion: Deep-tested. L2 outcome: tools_ok.

Free, no auth. Daily CN AI infra: LLM prices, index, trends, GPU, estimators, agents, PC, funding.

Add to your agent

Copy this into your MCP client config:

{
  "mcpServers": {
    "TANDIU AI infrastructure data": {
      "url": "https://www.tandiu.cn/mcp",
      "transport": "streamable-http"
    }
  }
}

Measured tools (14)

• about_tandiu_data read-only

  What data TANDIU publishes, refresh cadence, price basis and how to cite. Call first.

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• list_llm_prices read-only

  Official API list prices (USD per 1M tokens) with Artificial Analysis, LiveBench 2.0 and Epoch AI scores for major US and Chinese LLMs. Filter and sort.

  Parameters:

  • limit — integer
  • query — string — substring of model name
  • vendor — string
  • country — string — US, CN or another ISO code
  • sort_by — string

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• cheapest_model_for_score read-only

  Cheapest models whose benchmark score meets a threshold (absolute or percent of the best score).

  Parameters:

  • limit — integer
  • metric — string
  • country — string
  • min_score — number
  • min_percent_of_top — number

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• get_price_index read-only

  Chain-linked LLM price index (CN, US or all models), 30-day change, median price and series.

  Parameters:

  • group — string
  • since — string — YYYY-MM-DD

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• get_llm_price_history read-only

  Pro (API key required): daily official list-price history per model since July 2026, USD and RMB per 1M tokens, with source URLs. Filter by model, vendor, country and date range.

  Parameters:

  • limit — integer
  • model — string — model key or name substring, e.g. gpt-5.6, deepseek
  • since — string — YYYY-MM-DD
  • until — string — YYYY-MM-DD
  • vendor — string
  • country — string

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• list_price_changes read-only

  Official list-price cuts/increases observed in daily snapshots, and models newly added to tracking.

  Parameters:

  • limit — integer
  • since — string — YYYY-MM-DD, default last 90 days
  • min_change_pct — number

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• estimate_self_hosting read-only

  VRAM, GPUs needed, single-stream and batched throughput, hardware cost and the vendor's official API price/speed for self-hosting an open-weights model. Omit model to list available models.

  Parameters:

  • gpus — array — gpu ids from list_gpu_prices, e.g. h100_sxm, h20, ascend910b, rtx_pro_6000, rtx5090, rtx5090d
  • model — string

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• list_gpu_prices read-only

  Data-center, workstation and consumer GPUs (B300/B200/H200/H100/H20, Ascend 910B/910C, RTX PRO 6000/5500/5000, RTX 5090/5090D/4090D, L40S/L20): VRAM, bandwidth, FP16 compute, list price, China market price (SXM GPUs per 8-GPU node), cloud rental with 7-day change, sources and dates.

  Parameters:

  • query — string — gpu id or name substring, e.g. h20, rtx pro 6000, 910
  • with_cloud_rental — boolean
  • china_available_only — boolean

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• estimate_role_token_budget read-only

  Monthly token usage and cost for 32 common job roles at light / standard / heavy AI usage, with the best model (any country) and the best Chinese model that meet each role's capability floor.

  Parameters:

  • group — string
  • limit — integer
  • query — string — role id or name, e.g. backend, lawyer, doctor
  • usage — string

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• estimate_enterprise_ai_cost read-only

  Monthly AI cost for 10 / 100 / 1,000 / 10,000-person company samples across industries: API (per-role model choice, or the open model's official API) vs self-hosting the same open model on public cloud, private cloud or on-premises GPUs, with GPU count and cost per 1M tokens.

  Parameters:

  • query — string — industry or sample, e.g. bank, hospital, ecommerce
  • headcount — integer
  • include_team — boolean
  • include_gpu_breakdown — boolean — self-hosting cost on every GPU for each sample
  • include_hardware_table — boolean — per-GPU unit economics (throughput, hardware price, cost per 1M tokens)

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• list_cloud_agents read-only

  Cloud agents (one persistent agent per person, team agents, hosting platforms) with specs and monthly prices.

  Parameters:

  • kind — string
  • limit — integer
  • query — string

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• recommend_coding_pc read-only

  Relative speed and tier (best/recommended/minimum/not_advised) of Macs, laptops and desktops for running Codex / Claude Code.

  Parameters:

  • tier — string
  • query — string
  • category — string

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• list_ai_funding read-only

  Funding rounds of AI companies (mainly US and China): company, business, round, amount, valuation, investors, source links. Newest first; filter by region, sector, company/investor name, date or minimum amount.

  Parameters:

  • limit — integer
  • query — string — company, product or investor name
  • since — string — YYYY-MM-DD
  • region — string
  • sector — string
  • sort_by — string
  • min_amount_usd — number

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• list_overseas_listings read-only

  Where Chinese open models (DeepSeek, Qwen, GLM, Kimi, MiniMax, ...) are listed on platforms headquartered outside mainland China: hosting type, regions, input/output/blended USD per 1M tokens vs the official price, context, tool calling, data-policy links. Labs ranked by Hugging Face followers, models by downloads and likes.

  Parameters:

  • limit — integer
  • model — string — model name or key substring, e.g. deepseek v4 pro, kimi, glm-5
  • platform — string — platform name substring, e.g. together, bedrock
  • tool_calling — boolean
  • max_blended_usd — number — max blended USD per 1M tokens

  Hints: read-only yes · destructive no · idempotent yes · open-world no

Publisher website

Measured by MCPYP probing. Untested fields are omitted, never estimated. Listing is always free; MCPYP never proxies calls or stores user tokens.