Sciverse 工具

2. Install the SDK

The official and correct package name for both pip and npm is sciverse.

```bash

Quickstart — Skills CLI

The easiest way to install the skill for projects supporting the Skills CLI is via the npx skills command:

npx skills add https://sciverse.space

This command automatically fetches the skill manifest and registers the tool for your project. Don't forget to configure your API token via the SCIVERSE_API_TOKEN environment variable.

Quickstart — Claude Code

claude /plugin marketplace add https://github.com/opendatalab/Sciverse-Agent-Tools
claude /plugin install sciverse

The skill depends on sciverse-mcp-server; install it once:

npm install -g sciverse-mcp-server
export SCIVERSE_API_TOKEN=sv-...     # get one from https://sciverse.space

Or declare the MCP server per-project — see skill-claude-code/SKILL.md.

Quickstart — other MCP-capable agents

Drop this snippet into your agent's MCP config (.mcp.json for Claude Code / Cursor, ~/.codex/config.toml for Codex CLI, etc.):

{
  "mcpServers": {
    "sciverse": {
      "command": "npx",
      "args": ["-y", "sciverse-mcp-server"],
      "env": { "SCIVERSE_API_TOKEN": "${SCIVERSE_API_TOKEN}" }
    }
  }
}

Per-agent step-by-step guides:

Quickstart — agent host via well-known URL

For agent hosts that auto-discover skills via the well-known URI convention, Sciverse serves the skill bundle at:

https://sciverse.space/.well-known/agent-skills/index.json

The endpoint returns a manifest listing the sciverse skill and its files (SKILL.md, references, agent adapter configs, runnable scripts). Hosts that follow the convention fetch the manifest, then materialise the skill locally for the model to invoke.

Use this channel when:

• Your agent host already supports .well-known/agent-skills/ discovery
• You want the latest skill version automatically (no version pinning on the consumer side)
• You don't want to clone a git repo just to obtain the skill

For host-specific install commands (Claude Code, MCP, OpenClaw, ClawHub), see the other Quickstart sections above.

Quickstart — SDK

3. Configure credentials (any one of the three)

```bash

A. Environment variable (recommended for servers / CI)

export SCIVERSE_API_TOKEN=sv-...

4. Call the SDK

Python:

import asyncio
from sciverse import AgentToolsClient

async def main():
    # token / base_url omitted — resolved from env or credentials file
    async with AgentToolsClient() as c:
        r = await c.semantic_search(query="Transformer attention mechanism")
        for hit in r["hits"][:3]:
            print(hit["title"], hit["score"])

asyncio.run(main())

TypeScript:

import { AgentToolsClient } from "sciverse";

const c = new AgentToolsClient();  // reads SCIVERSE_API_TOKEN from env
const r: any = await c.semanticSearch({ query: "Transformer attention mechanism" });
r.hits.slice(0, 3).forEach((h: any) => console.log(h.title, h.score));

API at a glance

Python SDK

async with AgentToolsClient() as c:           # token from env / credentials file
    # 1. Field discovery — call once when first integrating
    await c.list_catalog(include_sample_values=True)
    # 2. Structured search
    await c.search_papers(query=..., authors=[...], year_from=2020, page_size=10)
    # 3. Semantic search (mode: fast / balanced / quality)
    await c.semantic_search(query=..., top_k=10, mode="balanced")
    # 4. Byte-range read of original content
    await c.read_content(doc_id=..., offset=0, limit=4096)
    # 5. Figure / table image bytes (multimodal RAG)
    img_bytes, mime = await c.get_resource(file_name="dt=.../p_.../f3.png")

Return values are typed as dict[str, Any]. The full response schema lives in openapi.yaml. Advanced users can from sciverse.types import SearchPapersRequest, ... for typed construction and validation.

Long-lived client (web server, agent runtime — outlives a single request):

client = AgentToolsClient()
try:
    while serving:
        r = await client.semantic_search(query=...)
        ...
finally:
    await client.aclose()   # close underlying httpx connection pool

TypeScript SDK

const c = new AgentToolsClient();   // token from env
await c.listCatalog({ include_sample_values: true });
await c.searchPapers({ query, authors, year_from, page_size });
await c.semanticSearch({ query, top_k, mode });
await c.readContent({ doc_id, offset, limit });
const { bytes, mimeType } = await c.getResource({ file_name });

Return values are typed as unknown — cast them yourself:

import type { components } from "sciverse";
type SemanticSearchResp = components["schemas"]["SemanticSearchResponse"];
const r = await c.semanticSearch({ query: "x" }) as SemanticSearchResp;

Pick your integration path