CostBench MCP Server
Enables access to verified software-pricing intelligence for over 3,290 products, including true costs, hidden fees, negotiation data, price history, and TCO calculations, all sourced and dated.
README
CostBench MCP Server
Verified software-pricing intelligence for any MCP client — true costs, hidden fees, negotiation data, price history, side-by-side comparisons, and total cost of ownership for 3,290+ products. Read-only, free, no API key, no signup.
Powered by CostBench.com. Every figure the server returns carries its source and date — the data is only as trustworthy as its provenance, so CostBench publishes the provenance with each number. Responses include an attribution field and per-entity links back to the sourced page.
Tools
| Tool | Description |
|---|---|
get_pricing |
Full sourced pricing record for a product: tiers, hidden costs, contract terms, market data, sentiment, negotiation tips, TCO scenarios, and price history — each claim with its sources. Args: slug, optional depth (summary|full). |
get_hidden_costs |
The documented hidden / implementation / add-on costs (uplifts, onboarding fees, overages), each with a sourced quote + date. Args: slug. |
get_negotiation_tips |
Sourced negotiation tactics with success-likelihood. Args: slug. |
get_price_history |
Per-quarter price snapshots + trend — has the price changed? Args: slug. |
compare_software |
Side-by-side pricing for 2–4 products (starting price, median annual cost, free tier, ratings). Args: slugs (array). |
discover_software |
Find products by category, maxPrice, hasFreeTier. Returns matching slugs to look up. |
calculate_tco |
Total cost of ownership for a product and team size: sourced line items, year-one + multi-year totals, effective per-seat cost. Args: slug, seats, optional tier, termYears. |
estimate_llm_cost |
Cost of an LLM/API workload (input + output tokens) for a usage-priced provider, from its real rate card. Args: slug, inputTokens, outputTokens, optional model. |
All tools are read-only (readOnlyHint: true). Products are addressed by slug (e.g. salesforce, microsoft-teams, openai-api); use discover_software if you don't know one.
Option 1 — Remote endpoint (recommended, no install)
The server is hosted at https://costbench.com/mcp using streamable HTTP transport (MCP spec 2025-03-26). If your client supports remote MCP servers, point it straight at the endpoint:
{
"mcpServers": {
"costbench": {
"type": "streamable-http",
"url": "https://costbench.com/mcp"
}
}
}
Claude Code:
claude mcp add --transport http costbench https://costbench.com/mcp
Option 2 — Local stdio server (this package)
For clients that only speak stdio, this package bridges stdio ↔ the hosted endpoint:
{
"mcpServers": {
"costbench": {
"command": "npx",
"args": ["-y", "github:aadilr/costbench-mcp"]
}
}
}
Or clone and run directly:
git clone https://github.com/aadilr/costbench-mcp.git
cd costbench-mcp
npm install
node index.js
Option 3 — Agent Skill / Claude Code plugin
This repo doubles as an Agent Skill and Claude Code plugin.
Any skills-capable agent (Claude Code, Codex CLI, Cursor, Gemini CLI, Copilot, and more):
npx skills add aadilr/costbench-mcp
Claude Code plugin (bundles the MCP server + the skill):
/plugin marketplace add aadilr/costbench-mcp
/plugin install costbench@costbench
The software-pricing skill prefers the MCP tools when connected and otherwise falls back to a bundled script (skills/software-pricing/scripts/pricing.sh) that talks to the hosted endpoint over plain HTTPS — no MCP client required.
Option 4 — Docker
docker build -t costbench-mcp .
docker run -i --rm costbench-mcp
Quick test
# List the tools
curl -X POST https://costbench.com/mcp \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'
# Get pricing for a product
curl -X POST https://costbench.com/mcp \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"get_pricing","arguments":{"slug":"salesforce","depth":"summary"}}}'
The get_pricing call above returns a record like:
{
"slug": "salesforce",
"name": "Salesforce",
"category": { "slug": "crm", "name": "CRM" },
"priceRange": { "low": 0, "median": 100, "high": 550, "currency": "USD" },
"priceUnit": "user/month",
"tiers": [
{ "name": "Starter Suite", "priceMonthly": 25, "priceAnnual": 300, "priceUnit": "user/month" },
{ "name": "Pro Suite", "priceMonthly": 100, "priceAnnual": 1200, "priceUnit": "user/month" }
],
"attribution": "Source: CostBench (https://costbench.com) — verified, sourced software pricing intelligence."
}
What's in the data
| Products | 3,290+ |
| Categories | 255 |
| Comparisons | 5,850+ |
| Coverage | Tiers, hidden costs, contract terms, negotiation data, price history, TCO, LLM rate cards |
Each field carries a license marker (owned | restricted): CostBench's own sourced research is owned; a small number of third-party market-data medians are restricted — surfaceable to a user but not licensed for commercial redistribution.
Environment variables
| Variable | Default | Purpose |
|---|---|---|
COSTBENCH_MCP_URL |
https://costbench.com/mcp |
Override the upstream endpoint |
Privacy & limits
- No authentication or account required. All tools are read-only — the server never writes anything.
- Per-IP rate limiting: 60 requests/minute.
- Data is sourced and dated; cite the
links.pageURL a response returns, not the bare number.
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。