Sapiom MCP Server
一個 API 金鑰即可讓代理程式存取 80 多種工具：網路搜尋、深度搜尋、瀏覽器自動化、螢幕截圖、400 多種 LLM 模型、圖片生成、文字轉語音、音效及電話驗證。按用量計費，內建支出管理功能。
文件
Sapiom
All You Need
MCP ServerSkill
$claude mcp add sapiom -- npx -y @sapiom/mcp
sapiom-prompt
Scan every AI paper published this week, build a knowledge graph of how they connect, and deploy a searchable interface my team can explore visually.
Instant access to 20+ paid services with pay-per-use pricing. No vendor account setup required.
Two ways in
Sapiom ships two MCP servers, for two jobs — plus a plain SDK:
- Authoring MCP (
@sapiom/mcp, local vianpx) — build, test & deploy workflows. Inside a step, capabilities are called via the typedctx.sapiom.*client. - Remote MCP (
https://api.sapiom.ai/v1/mcp) — call any capability directly as ansapiom_*tool, no workflow needed. - SDK (
@sapiom/axios/@sapiom/fetch) — wrap your HTTP client and hit the gateways from code.
Rule of thumb: authoring MCP for anything multi-step, scheduled, or deployable; remote MCP (or the SDK) for a one-off call.
Authoring MCP — build workflows
Connect your AI coding agent to the Sapiom authoring MCP (@sapiom/mcp):
Claude Code:
claude mcp add sapiom -- npx -y @sapiom/mcp
Other clients (Cursor, Windsurf, etc.):
{
"mcpServers": {
"sapiom": {
"command": "npx",
"args": ["-y", "@sapiom/mcp"]
}
}
}
Sign in once with the sapiom_authenticate tool — it opens a browser to approve access, then caches your key in ~/.sapiom/credentials.json for every subsequent call. No API key to paste.
Remote MCP — call capabilities directly
Connect to the hosted remote MCP to call any capability as a direct sapiom_* tool:
Claude Code:
claude mcp add sapiom --transport http https://api.sapiom.ai/v1/mcp
{
"sapiom": {
"transport": "http",
"url": "https://api.sapiom.ai/v1/mcp"
}
}
}
If your platform requires explicit auth, add --header "x-api-key: YOUR_API_KEY" (CLI) or a "headers": { "x-api-key": "YOUR_API_KEY" } block (JSON). Get a key at app.sapiom.ai/settings.
What You Can Do — direct tool calls (remote MCP)
These sapiom_* tools are exposed by the remote MCP for one-off, direct calls. Inside a workflow, call the same capabilities via ctx.sapiom.* — see Workflows.
| Category | Examples |
|---|---|
| Search | Web search with AI answers (Linkup), web + news results (You.com) |
| Browser | Fetch page content as markdown, capture screenshots |
| Audio | Text-to-speech, sound effects, voice listing |
| AI Models | Chat with LLMs (GPT-4o, Claude, etc.), list models and pricing |
| Image | Generate images from text (FLUX Schnell/Dev/Pro) |
| Verify | SMS verification (send + check codes) |
| Governance | Agent registration, spending rules, scoped API keys |
Use tool_discover (remote MCP) to find the right tools for your task — describe a goal and get ranked recommendations with examples.
Recipes
Chain remote-MCP tools for common one-off tasks:
Research Report
sapiom_search → sapiom_deep_search → sapiom_chat
Search both Linkup and You.com for broad source coverage, then synthesize findings into a structured report using an LLM. Use depth: "deep" on sapiom_search for thorough coverage of complex topics.
Visual Website Audit
sapiom_screenshot → sapiom_chat → sapiom_generate_image
Capture a screenshot of a live website, analyze the design and UX with an LLM, then generate a mockup image of proposed improvements.
Podcast Episode
sapiom_search → sapiom_chat → sapiom_text_to_speech
Research a topic with web search, write a podcast script with an LLM, then convert to audio. Use sapiom_list_voices to pick a voice. Split scripts over ~400 words into chunks.
Governed Agent Workflow
sapiom_create_agent → sapiom_create_spending_rule → sapiom_create_transaction_api_key
Register a named agent, set spending limits (e.g., $100/day rolling cap), and create a scoped API key for autonomous operation. Use agentName on subsequent tool calls for spend tracking.
Competitor Intelligence
sapiom_fetch → sapiom_deep_search → sapiom_chat → sapiom_generate_image
Extract content from a competitor’s website, search for recent news and coverage, synthesize an analysis with an LLM, and generate a comparison infographic.
Tips
- Start with
tool_discover— it recommends tools based on your objective and returns usage examples - Use
agentNameon tool calls to track spend per agent and enforce spending rules - Free endpoints —
sapiom_status,sapiom_list_voices, andsapiom_list_modelscost nothing - Combine search providers —
sapiom_search(Linkup) andsapiom_deep_search(You.com) use different indexes for broader coverage - Spending rules — use
sapiom_create_spending_ruleto cap costs by amount, count, or time window
SDK Access (Direct HTTP)
For programmatic access without MCP, use the Node.js SDK:
import { withSapiom } from "@sapiom/axios";
import axios from "axios";
const client = withSapiom(axios.create(), {
apiKey: process.env.SAPIOM_API_KEY,
});
const { data } = await client.post(
"https://linkup.services.sapiom.ai/v1/search",
{ query: "quantum computing", depth: "standard", outputType: "sourcedAnswer" }
);
npm install @sapiom/axios axios
Python and other languages: use any HTTP client with Authorization: Bearer sk_... header against gateway URLs.
Documentation
Fetch any doc as markdown by adding .md to the URL.