WebMCP Tools for AI Pricing and Hardware Data
Read-only WebMCP tools that let AI agents query RAM, SSD and GPU prices, AI model pricing and benchmarks, with a source and a date on every answer.
Capital & Compute registers 12 read-only WebMCP tools on every page. WebMCP is a draft browser API from the W3C Web Machine Learning Community Group that lets a website hand an AI agent typed tools to call, instead of leaving the agent to read the page or click through it. An agent visiting this site can ask for the current DDR5 kit price or the cheapest model that clears a benchmark and get structured data back, with the page it came from attached.
What every answer includes
Each data tool answers from the same build-time datasets the human pages render, so a tool never says something the page does not. Every response carries:
- canonical_url: the page the figures belong to, to link when you cite them.
- as_of: the date the dataset was last verified.
- cite_as: a ready-made attribution line.
- Primary sources: row-level source URLs, or the tracker source list, so a decisive number can be checked against the original.
All tools are marked readOnlyHint. None of them submit forms, sign anyone up, or change anything.
Why the answers come with notes attached
The hardware trackers mix price channels that do not compare: a TrendForce contract move, a wafer spot reading, an Amazon retail listing and a reseller street price can all describe "the price of memory" in the same week and point in different directions. Retail rows are stored as ranges because the two free retail trackers disagree on the level. A tool that flattened those into a single figure would hand an agent a cleaner number that is also a wrong one. So each tracker block comes back with the note that explains its channel and its limits, and the tracker overview lists every block with a one-line description before the agent asks for one. The caveat travels with the figure into whatever answer the agent writes.
The tools
| Tool | What it returns | Input | Backs |
|---|---|---|---|
| get_page_summary | Summarize the Capital & Compute page currently open: title, description, publish and update dates, section outline and canonical URL. Use it to cite this page accurately. | none | Any page |
| list_articles | List every Capital & Compute post, data tracker and tool with a one-line summary and URL (the machine-readable /llms.txt catalog). Covers AI model pricing, coding agents, AI benchmarks, and RAM, SSD and GPU prices. | none | Capital & Compute catalog |
| get_memory_prices | Current RAM and DRAM prices with sources: DDR4 spot price history, DDR5 retail kit prices by capacity, server RDIMM and HBM prices, and memory price indexes. Updated as the memory shortage moves. Call with no section for an overview. | section | RAM Price Index |
| get_ssd_prices | SSD and NAND flash prices with sources: consumer cost per TB by capacity, enterprise NVMe street prices, hard drive prices, and TrendForce NAND contract price moves. Call with no section for an overview. | section | SSD Price Tracker |
| get_gpu_prices | Graphics card prices: Nvidia RTX 50 and AMD Radeon RX official MSRP against street price, the 2026 GPU price hikes, and GDDR memory cost per card. Call with no section for an overview. | section | GPU Price Tracker |
| get_hosting_price_hikes | Cloud and dedicated server price increases in 2026 (Hetzner, OVHcloud and others), with dates, percentages and the memory-cost reasons providers gave. Call with no section for an overview. | section | Hosting Price Hikes tracker |
| get_ai_model_leaderboard | AI model value leaderboard: verified API price per million tokens, coding and intelligence benchmark composites, and value per dollar for current LLMs (Claude, GPT, Gemini, DeepSeek, Qwen, Kimi and others). | query, limit | AI Model Value Leaderboard |
| get_recent_model_releases | AI model release record by month: release date, provider, launch API price per million tokens, context window, open-weight status and a one-line summary of each notable LLM launch. | query, limit | AI Model Releases by Month |
| find_ai_benchmark | Look up an AI benchmark (SWE-bench, GPQA, ARC-AGI, Humanity’s Last Exam and 110+ others): what it measures, who made it, its metric, whether it is saturated, and a representative top score with sources. | query, limit | AI Benchmarks directory |
| find_inference_providers | Compare AI inference API providers (OpenAI, Anthropic, Google, Together, Fireworks, Groq, Cerebras, OpenRouter and others): pricing model, representative prices, flagship model, free tier and OpenAI compatibility. | query, limit | AI Inference Providers directory |
| list_free_ai_models | Good AI models you can call through an API for free, ranked by the Artificial Analysis Intelligence Index, with the free hosts that serve them and their limits. | query, limit | Best Free AI Model APIs |
| find_claude_skill | Search a curated directory of Claude Skills, agent skills, MCP servers and prompt libraries: what each does, who maintains it, how to install it, and its source. | query, limit | Best Claude Skills, Agent Skills & MCP Servers |
Using them without a browser agent
The payloads the data tools fetch are plain static JSON, so any agent or script can read them directly. Tracker tools also serve one file per block at /webmcp/<tool>/<section>.json.
/webmcp/get_memory_prices.json: RAM Price Index/webmcp/get_ssd_prices.json: SSD Price Tracker/webmcp/get_gpu_prices.json: GPU Price Tracker/webmcp/get_hosting_price_hikes.json: Hosting Price Hikes tracker/webmcp/get_ai_model_leaderboard.json: AI Model Value Leaderboard/webmcp/get_recent_model_releases.json: AI Model Releases by Month/webmcp/find_ai_benchmark.json: AI Benchmarks directory/webmcp/find_inference_providers.json: AI Inference Providers directory/webmcp/list_free_ai_models.json: Best Free AI Model APIs/webmcp/find_claude_skill.json: Best Claude Skills, Agent Skills & MCP Servers
The full datasets behind the trackers are published too, and the site catalog lives at /llms.txt:
/memory-prices.json: RAM Price Index/ssd-prices.json: SSD Price Tracker/hosting-price-hikes.json: Hosting Price Hikes/ai-model-leaderboard.json: AI Model Value Leaderboard/ai-model-releases.json: AI Model Releases/ai-benchmarks.json: AI Benchmarks/ai-inference-providers.json: AI Inference Providers/free-ai-models.json: Free AI Models/claude-skills.json: Claude Skills directory
Agent Skills describing how to use the pricing data are at /.well-known/agent-skills/.
Trying the tools in Chrome
WebMCP is still experimental. According to Chrome's WebMCP documentation, you can turn it on locally with the flag chrome://flags/#enable-webmcp-testing, then open any page here: the tools register as soon as document.modelContext is available. For ordinary visitors the tools cost nothing: the page checks whether the browser supports WebMCP and loads the tool module only if it does.
Whether WebMCP actually saves agents money is a separate question, and most of the numbers passed around about it do not hold up. WebMCP token savings: which numbers are real checks each claim against the published test data.
Corrections and requests
If a tool returns a stale or wrong figure, or you want a dataset exposed as a tool, use the contact page. Data corrections follow the editorial standards.