Connect Sciverse Skill in Claude / Cursor / Codex

Install Sciverse MCP tools so AI assistants can retrieve scientific literature directly

Scenario

Developers want to call Sciverse from AI coding assistants such as Claude Code, Cursor, or Codex CLI without writing API request code manually.

Estimated calls

~2-5 tool calls per conversation

Tools

Sciverse MCP Serversemantic_searchread_contentlist_catalog

Pipeline

安装 Skill / 配置 MCP→ 设置 Token→ AI 助手自动调用→ 输出带引用的证据

Input example

User asks the AI assistant:
"Find recent papers on protein language models and summarize the key evidence."

Output example

The assistant calls Sciverse tools, retrieves papers, reads source snippets, and returns a cited answer in the conversation.

Agent Prompt example

You are an AI assistant with Sciverse tools installed. For scientific literature questions, call the relevant Sciverse tool first, then answer with citations and source metadata.

Implementation steps

Step 1: Install Sciverse Skill

Install the Sciverse skill or MCP integration in your assistant

# ─── Claude Code ───
# 通过官方域名安装（推荐）
npx skills add https://sciverse.space

# ─── Codex CLI ───
# 参考 Codex 文档配置 MCP server

Step 2: Configure API token

Set SCIVERSE_API_TOKEN in the environment used by the assistant

# 在 shell 配置文件中添加（~/.bashrc 或 ~/.zshrc）
export SCIVERSE_API_TOKEN="sv-your-token-here"

# 或在项目 .env 文件中
SCIVERSE_API_TOKEN=sv-your-token-here

# 验证安装成功（Claude Code）
npx skills list | grep sciverse

Step 3: Use it in an AI assistant

Ask scientific questions and let the assistant call Sciverse tools

# 在 Claude Code / Cursor / Codex 中直接提问：

> 帮我查找 2023 年以来关于 LLM 幻觉检测的论文

AI 助手会自动：
1. 调用 semantic_search(query="LLM hallucination detection", top_k=10)
2. 返回相关论文片段和引用
3. 如需详情，继续调用 read_content 获取全文

> 用结构化检索按 Nature 期刊过滤

AI 助手会：
1. 调用 list_catalog() 确认字段名
2. 调用 search_papers(filters=[...]) 执行过滤

Notes

The assistant process must receive SCIVERSE_API_TOKEN.
Use the official Sciverse Agent Tools repository as the source of truth.
Tool availability depends on the client and installation method.
Start with simple retrieval prompts before building multi-step workflows.

FAQ

这个案例适合谁?

适合把 Sciverse 封装进 Cursor、Claude、Codex 或自研 Agent 工具链的开发者。

建议封装哪些工具?

至少封装检索、读取全文、获取资源和元数据筛选四类工具。

工具描述要写什么?

要写清楚输入字段、返回字段、适用场景和不适用边界。

如何减少 Agent 调错接口?

在工具说明中明确 meta-search、agentic-search、content、resource 的边界。

Next steps

Need an API key?

Create one in Console > Tokens.The same API key works for enabled Sciverse, DianShi, and Skills capabilities, with starter quota available according to account permissions.

Open console