Sciverse Skills

Based on the opendatalab/Sciverse-Agent-Tools repository, Sciverse provides 5 standardized agent tools plus Python / TypeScript SDKs, with 5 setup paths.

GitHub

opendatalab/Sciverse-Agent-Tools

list_catalog

Field catalog

search_papers

Structured search

semantic_search

Semantic search

read_content

Text slices

get_resource

Figures/resources

Setup paths

npx skills add

npx skills add https://sciverse.space

OpenClaw · ClawHub

clawhub install sciverse

Claude Plugin

/plugin install sciverse

Manual Skill

clone then copy to .claude/skills

Python · TS SDK

pip / npm install sciverse

Compatible with Claude Code, Cursor, Codex CLI, Windsurf, and other mainstream agent clients through sciverse-mcp-server.

Connection

Transport: sciverse-mcp-server / SDK / ClawHub Skill
Endpoint: https://github.com/opendatalab/Sciverse-Agent-Tools
Auth: Use a Sciverse API token (SCIVERSE_API_TOKEN environment variable). Fill it in the panel for one-click ClawHub setup; Claude Code, Cursor, Codex CLI, Windsurf, and others connect through sciverse-mcp-server.

Client config

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

Setup methods

# One-command Agent Skill install (recommended for environments that support npx skills)

# 1) Install from the official source
npx skills add https://sciverse.space

# 2) Set the token
export SCIVERSE_API_TOKEN=sv-xxx

Tools overview

Category	Tool	Description	Typical latency
SDK	list_catalog	List available fields, filter operators, and sample values; call once before first integration or before composing precise filters.	~120ms
SDK	search_papers	Structured metadata search (author / year / journal / subject), useful for precise filtering before semantic retrieval.	~150–500ms
SDK	semantic_search	Semantic search over literature snippets using natural language; the main entry point for RAG scenarios.	~500–1500ms
SDK	read_content	Read source text by byte range to expand RAG context, using doc_id / offset returned by semantic_search.	~200ms
SDK	get_resource	Fetch figure/table binaries referenced in read_content Markdown (image bytes + MIME) for multimodal RAG.	~200ms

SDK tools

list_catalog

~120ms

List available fields, filter operators, and sample values; call once before first integration or before composing precise filters.

Field	Type	Required	Description
include_sample_values	boolean	Optional	Whether to return sample values with enum fields for LLM self-checking. Default false

Returns: fields[]（name · type · operators[] · sample_values?）

search_papers

~150–500ms

Structured metadata search (author / year / journal / subject), useful for precise filtering before semantic retrieval.

Field	Type	Required	Description
query	string	Optional	Additional natural-language terms.
authors	string[]	Optional	Author name list.
year_from	integer	Optional	Start year.
filters_advanced	object[]	Optional	Precise filter expression composed from fields returned by list_catalog.
page_size	integer	Optional	Items per page. Default 10Range 1–100

Returns: hits[]（title · doc_id · authors · year · venue · abstract）

semantic_search

~500–1500ms

Semantic search over literature snippets using natural language; the main entry point for RAG scenarios.

Field	Type	Required	Description
query	string	Required	Natural-language query, 1-2048 characters.
top_k	integer	Optional	Maximum snippets to return. fast/balanced effectively cap at ~50; quality can exceed that. Default 10Range 1–100
mode	string	Optional	Quality / latency tradeoff for semantic retrieval. Default balancedRange fast / balanced / quality

Returns: hits[]（doc_id · chunk_id · text · score · source.title · source.year · offset）

read_content

~200ms

Read source text by byte range to expand RAG context, using doc_id / offset returned by semantic_search.

Field	Type	Required	Description
doc_id	string	Required	Document ID returned by search_papers / semantic_search.
offset	integer	Optional	Starting byte offset. Default 0
limit	integer	Optional	Maximum bytes to fetch in this request. Default 4096

Returns: Markdown snippet (with ![](dt=.../p_.../f*.png) references) · next_offset · more

get_resource

~200ms

Fetch figure/table binaries referenced in read_content Markdown (image bytes + MIME) for multimodal RAG.

Field	Type	Required	Description
file_name	string	Required	Path such as dt=xxx/p_yyy/f3.png, provided in Markdown returned by read_content.

Returns: bytes (image/png, etc.) · mime_type

Error response

{
  "content": [{ "type": "text", "text": "Error message" }],
  "isError": true
}

Scenario	HTTP status	Description
Token missing or invalid	401	Check whether SCIVERSE_API_TOKEN is configured in the client or environment.
Invalid request parameters	400	Raised as Python httpx.HTTPStatusError / TypeScript Error("Sciverse API 400: ...").
Quota exceeded	429	Returned by the service only; wait for the limit window or contact support to adjust account quota.
Service temporarily unavailable	502 / 503	Retry with exponential backoff (1s / 2s / 4s).

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