Universal

You can access OpenAI vector stores and manage files, uploads, and batch operations through a production-ready MCP server that supports multiple deployment options. This server lets you run in the cloud, locally via stdio, or as a development server, and it provides a complete, type-safe interface for vector store workflows used by AI assistants and MCP clients.

How to use

Connect to the MCP server using your preferred client (Claude Desktop, Roo, or other MCP-compatible clients) and perform end-to-end vector store workflows. You can upload local files, create vector stores, add files to stores, and run batch operations to process large collections. Authentication is handled via an API key passed through the MCP channel, and every action is executed against your OpenAI vector store resources.

Key capabilities you can perform include creating and listing vector stores, uploading files from your computer, adding files to stores, retrieving file contents, creating multipart uploads for large files, and managing batch operations that process multiple files at once. Use the available tools to build your document knowledge base, perform code or document analysis, and run queries against your stored documents.

How to install

# Prereqs: ensure Node.js and npm are installed if you plan to run the npm/stdio option
node -v
npm -v

Next choose one of the three deployment options and follow the corresponding setup steps shown in the sections below.

## Option 1. Cloudflare Workers (Production URL)

Use the production endpoint that runs on Cloudflare Workers for zero-setup, global edge distribution, and sub-100ms responses.

Access URL: https://vectorstore.jezweb.com/mcp (use your API key in the path as shown when authenticating)

## Option 2. NPM Package (Local Stdio)

Install and run the MCP server locally using the NPM package. This mode uses a direct stdio transport for fastest communication with your MCP client.

Install and run using the latest release via npx, or install globally or in your project as needed.

## Option 3. Local Development Server

Clone the project locally, install dependencies, and run the development server to customize and test your own implementation.

The development workflow typically uses npm run dev to start a local server, which you then configure in your MCP client.

## Available tools

### vector-store-create

Create a new vector store with optional expiration and metadata.

### vector-store-list

List all vector stores with pagination and sorting.

### vector-store-get

Get detailed information about a specific vector store.

### vector-store-delete

Delete a vector store permanently.

### vector-store-modify

Update vector store name, expiration, or metadata.

### file-upload

Upload local files directly to OpenAI for end-to-end workflows.

### file-list

List all uploaded files with filtering and pagination.

### file-get

Get detailed information about specific files.

### file-delete

Remove files from OpenAI storage.

### file-content

Download and retrieve file content.

### upload-create

Create multipart uploads for large files (>25MB).

### vector-store-file-add

Add an existing file to a vector store.

### vector-store-file-list

List all files in a vector store with filtering.

### vector-store-file-get

Get details of a specific file in a vector store.

### vector-store-file-content

Retrieve the content of a file in a vector store.

### vector-store-file-update

Update file metadata.

### vector-store-file-delete

Remove a file from a vector store.

### vector-store-file-batch-create

Create a batch operation for multiple files.

### vector-store-file-batch-get

Get the status of a batch operation.

### vector-store-file-batch-cancel

Cancel a running batch operation.

### vector-store-file-batch-files

List files in a batch operation.