bigapi
@bigapi/mcp
MCP 服务器,用于 bigapi.dev —— AI 代理的输出层。
为 Claude Desktop、Cursor、Cline、Windsurf 以及任何支持 MCP 的代理提供 LLM 自身无法完成的文件操作:
工具 | 功能 |
| HTML / Markdown / URL → PDF 或 PNG(报告、发票、报价、截图) |
| PDF 基础操作 |
| PDF 页面 → JPEG/PNG,例如用视觉模型查看文档 |
| 调整大小、裁剪、旋转、转换(webp/avif/…)、压缩、去除 EXIF、添加水印——一次调用 |
| 格式、尺寸、色彩空间、EXIF/ICC 是否存在 |
| 免费 API 密钥,即时获取,无需注册——100 次免费操作 |
| 账户 |
每次操作 1 美分。100 次免费。余额永不过期。失败的调用免费。 服务器位于德国,文件在交付后删除。
安装
需要 Node 18+。无需预先提供 API 密钥——代理可以自行调用 get_access。
Claude Desktop
claude_desktop_config.json(设置 → 开发者 → 编辑配置):
{
"mcpServers": {
"bigapi": {
"command": "npx",
"args": ["-y", "@bigapi/mcp"]
}
}
}重启 Claude Desktop。然后说:“获取 bigapi 访问权限,并将这段文本以 PDF 形式渲染到我的桌面。”
Cursor / Windsurf / Cline
将同样的配置块放入各自的 MCP 设置中(.cursor/mcp.json、~/.codeium/windsurf/mcp_config.json、Cline → MCP 服务器 → 配置)。
使用已有密钥
"bigapi": { "command": "npx", "args": ["-y", "@bigapi/mcp"], "env": { "BIGAPI_KEY": "bigapi_..." } }Related MCP server: Filesystem MCP Server
文件的工作方式
输入是本地路径(/Users/me/report.pdf、C:\Users\me\scan.pdf)。如果指定了 output_path,输出将写入该路径;否则写入临时文件夹(可通过 BIGAPI_OUTPUT_DIR 更改)。每个结果都包含费用、扣费来源以及剩余余额。
环境变量
变量 | 默认值 | 用途 |
| – | 使用此密钥代替已存储的密钥 |
|
|
|
| 操作系统临时目录 | 默认输出文件夹 |
|
| API 基础地址(用于自托管/测试) |
不使用 MCP
纯 HTTP 在任何地方都能使用(n8n、Make、Zapier、LangChain、你自己的代码):
curl -X POST https://api.bigapi.dev/v1/keys # → key
curl -o out.pdf https://api.bigapi.dev/v1/render \
-H "Authorization: Bearer $KEY" -H "content-type: application/json" \
-d '{"markdown":"# Hello from an agent"}'OpenAPI:https://api.bigapi.dev/openapi.json · 文档:https://api.bigapi.dev/docs · llms.txt:https://api.bigapi.dev/llms.txt
许可证
MIT
Available Tools
5 toolsenable_toolsA
Add the dedicated tools for specific operations to this session, e.g. ["pdf_redact","ocr"], or ["all"] for every operation. bigapi starts lean – only find_tool, run_operation, get_access and get_balance are listed – so your context stays free. You rarely need this: every operation already runs through run_operation. Reach for it when you will call the same operation many times and want its parameters spelled out in your tool list. Names come from find_tool; unknown names are reported back and skipped while the rest are still enabled. The effect lasts for this session, adds to what is already enabled, and cannot be undone from here – restart the server for the lean list again.
Inputs: names (required).
find_toolA
Find the right bigapi operation for a task. Describe what you need in plain words, English or German – "convert a png to webp", "remove customer names from a contract", "extract the ZUGFeRD invoice XML" – and get the matching operations with their parameters, a ready-to-run example, the price and a guide link. It recommends only: it never touches your files and never spends credit; run what it names with run_operation. Free and no API key needed, so it is also the cheapest way to see what bigapi covers. Start here whenever you are unsure which operation fits, and skip it when you already know the operation name. If nothing fits, the answer says so plainly instead of guessing.
Inputs: query (required), limit.
get_accessA
Get a bigapi API key for this machine. Call this once when no key is configured; other bigapi tools fail with "no API key" until you do. Creates a NEW free key (no signup, no credit card) and stores it in the local config file, where every bigapi tool picks it up. Calling it again creates an additional key rather than returning the existing one – use get_balance to check the key you already have. The new key includes 100 free operations, then $0.01 per operation from a prepaid balance; neither expires. Needs network access to api.bigapi.dev.
Inputs: name.
get_balanceA
Check the bigapi key that is currently configured: remaining credit, free operations left, monthly cap and spend so far this month. Read-only, free, and a snapshot of this moment – the numbers move as operations run. Use it to confirm a key works, before a large batch, or when an operation reports a low balance. It does not create keys (that is get_access) and does not list past operations.
Inputs: no named parameters.
run_operationA
Run any bigapi operation on real files: merge or redact a PDF, OCR a scan, convert an image, chunk text for embeddings, read a ZUGFeRD invoice, sign an image as AI-generated. Take the operation name from find_tool (e.g. "pdf/merge", "ocr", "text/chunk"). Uploads are local file paths. A file result is written to output_path, or to a temporary file when you omit it, and the path comes back with size and content type; a data result comes back as JSON. $0.01 per operation flat, no subscription, and failed calls cost nothing. This one executor covers every operation, including ones added after your client started – enable_tools only adds convenience wrappers around it. Files are processed in Germany, deleted right after delivery, and every operation ships with a published proof that it does what it promises.
Inputs: op (required), files, params, file_field, +1 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 5 tools
Each tool has a distinct lifecycle role: discovery (find_tool), execution (run_operation), session convenience (enable_tools), key creation (get_access), and balance checking (get_balance). The descriptions explicitly cross-reference each other and state what they do not do, so there is little risk of an agent picking the wrong one.
All five names follow a consistent snake_case verb_noun pattern (find_tool, run_operation, enable_tools, get_access, get_balance). The verbs are specific and match the action, and no style mixing or vague generic names appear.
Five tools is well-scoped for a server that intentionally stays lean and dynamically exposes operations through find_tool/run_operation. Each tool covers a necessary concern: discovery, execution, optional tool enabling, access, and balance.
The core workflow is fully covered: obtain a key, find an operation, run it, and check credit/usage. The dynamic design means new operations don't require new server tools, so there are no obvious dead ends or missing lifecycle steps.
Maintenance
Related MCP Connectors
MCP tools for AI agents: render URLs to image/PDF, check link health, convert HTML/CSV/JSON.
OCR, transcription, file extraction, and image generation for AI agents via MCP.
Generate images, GIFs, videos, and PDFs from HTML, URLs, or templates — from your AI agent.
Htmlpdf Transform Mcp connects AI agents to real public APIs via MCP. Tools include
Related MCP Servers
AlicenseBqualityCmaintenanceThis repository is an example of how to create a MCP server for Qdrant, a vector search engine.21,572Apache 2.0- FlicenseAqualityBmaintenanceNode.js server implementing Model Context Protocol (MCP) for filesystem operations.70141,222,611 npm92,139-

Chroma MCP Serverofficial
AlicenseAqualityDmaintenanceA server that provides data retrieval capabilities powered by Chroma embedding database, enabling AI models to create collections over generated data and user inputs, and retrieve that data using vector search, full text search, and metadata filtering.1386,671 PyPI605Apache 2.0
@rendershot/mcp-serverofficial
AlicenseAqualityDmaintenanceEnables AI agents to capture screenshots and generate PDFs from URLs or HTML via the Rendershot API.43MIT