Vesper REST API Skills

Vesper REST API Skills

A ready-to-use skill set that calls Vesper via REST API and returns structured JSON.

Environment

Self-hosted (recommended today): run the MCP server in HTTP mode (vespermcp --http, default port 8787), then point clients at http://127.0.0.1:8787 and use POST /v1/tools/call with the same tool names as MCP (see @vespermcp/mcp-server docs / docs/configuration.md).

Set these environment variables before using the skill script helpers (which still assume path-style endpoints unless you override them):

export VESPER_API_URL="https://your-vesper-host"
export VESPER_API_KEY="vesper_sk_..."

Auth header used for every request:

Authorization: Bearer VESPER_API_KEY

Skill Set

1. search_datasets - Find datasets by query.
2. download_dataset - Download dataset by ID.
3. clean_dataset - Run cleaning pipeline.
4. analyze_quality - Return quality report.
5. export_dataset - Export dataset to target format.

Endpoints

Default endpoint mapping:

• search_datasets -> /api/search_datasets
• download_dataset -> /api/download_dataset
• clean_dataset -> /api/clean_dataset
• analyze_quality -> /api/analyze_quality
• export_dataset -> /api/export_dataset

You can override each endpoint via env vars:

• VESPER_ENDPOINT_SEARCH_DATASETS
• VESPER_ENDPOINT_DOWNLOAD_DATASET
• VESPER_ENDPOINT_CLEAN_DATASET
• VESPER_ENDPOINT_ANALYZE_QUALITY
• VESPER_ENDPOINT_EXPORT_DATASET

File

Implementation is in vesper-rest-skills.js.

Usage

import {
  createVesperSkills,
  createOpenAIFunctions,
  createClaudeTools,
  createLangChainTools,
  createLangChainDynamicStructuredTools,
} from "./.agents/skills/vesper-rest-api/vesper-rest-skills.js";

const skills = createVesperSkills();
const search = skills.find((s) => s.name === "search_datasets");
const result = await search.invoke({ query: "medical imaging", limit: 5 });
console.log(result);

OpenAI Function Calling

const openAIFunctions = createOpenAIFunctions();

Returns:

• type: "function"
• function name
• function description
• JSON-schema parameters

Claude Tool Use

const claudeTools = createClaudeTools();

Returns:

• tool name
• tool description
• input_schema

LangChain Tools

const langchainTools = createLangChainTools();

Each tool object includes:

• name
• description
• Zod schema
• invoke(input) -> structured JSON

If you want native LangChain DynamicStructuredTool objects:

const dynamicTools = await createLangChainDynamicStructuredTools();

Structured JSON Output

All skill calls return a consistent JSON envelope:

{
  "skill": "search_datasets",
  "status": "success",
  "...": "endpoint-specific payload"
}

Source: sutaniese/Vesper — distributed by TomeVault.