VOC OpenAPI MCP Installer
Install the full VOC OpenAPI MCP server for Amazon review insights, feedback tagging, listing optimization, and hundreds of published data tools in local AI clients.
Built for e-commerce operators, consumer-insights teams, product researchers, and AI agents that need authorized review data and repeatable VOC workflows.
Usage
curl -fsSL https://open.voc.ai/install.sh | bash
By default, the installer opens your browser to sign in to VOC and approve access. After authorization, it automatically writes the MCP configuration for the supported clients detected on your machine, including Codex, Claude, Cursor, and others listed below.
Windows PowerShell:
irm https://open.voc.ai/install.ps1 | iex
Alternatively, use the npm package directly:
npx voc-openapi-mcp@latest setup
Codex users can also install the general-purpose VOC OpenAPI query skill:
npx voc-openapi-mcp@latest setup --with-skill
The skill is written to ~/.agents/skills/voc-openapi/SKILL.md only when Codex is detected. The uninstaller removes only installer-managed versions that have not been modified by the user.
By default, the installer registers the agents detected on the local machine. It currently supports:
- Claude Code (CLI; writes to
~/.claude.json) - Claude Desktop (desktop app; detects the macOS app or Windows
%LOCALAPPDATA%\AnthropicClaude, writes toclaude_desktop_config.json, and bridges the remote MCP server throughmcp-remote) - Cursor (desktop app or CLI launcher; detects the macOS app or Windows
%LOCALAPPDATA%\Programs\cursor, both using~/.cursor/mcp.json) - Windsurf (desktop app or CLI launcher; detects the macOS app or Windows
%LOCALAPPDATA%\Programs\Windsurf, both using~/.codeium/windsurf/mcp_config.json) - WorkBuddy (desktop app; writes user-level
~/.workbuddy/mcp.json) - Gemini CLI (writes to
~/.gemini/settings.json) - Codex (CLI and Desktop both use
~/.codex/config.toml)
List the agents detected on the local machine:
curl -fsSL https://open.voc.ai/install.sh | bash -s -- --list-agents
Windows PowerShell:
& ([scriptblock]::Create((irm https://open.voc.ai/install.ps1))) -ListAgents
Register every supported agent (the installer detects installed clients automatically by default, so this is usually unnecessary):
npx voc-openapi-mcp@latest setup --all-agents
Default configuration:
- Codex:
$CODEX_HOME/config.toml, or~/.codex/config.tomlwhenCODEX_HOMEis not set - Claude Code:
~/.claude.json - Claude Desktop:
~/Library/Application Support/Claude/claude_desktop_config.json(or%APPDATA%\Claude\claude_desktop_config.jsonon Windows) - Cursor:
~/.cursor/mcp.json - Windsurf:
~/.codeium/windsurf/mcp_config.json - WorkBuddy:
~/.workbuddy/mcp.json - Gemini CLI:
~/.gemini/settings.json - MCP server name:
voc-openapi - MCP URL:
https://open.voc.ai/mcp-server - Authentication header:
X-API-Key - Full endpoint: all published VOC OpenAPI MCP tools
- Codex MCP startup timeout:
120seconds
The installer updates the corresponding MCP server configuration in place and does not create .bak backup files.
Options
voc-mcp setup \
--api-base https://open.voc.ai \
--name voc-openapi \
--with-skill \
--config ~/.codex/config.toml
Authentication
Browser sign-in is the default and recommended flow. The installer opens the VOC authorization page in your default browser, waits for you to sign in and approve access, then updates the detected AI clients automatically. If the browser cannot be opened, the installer prints the same authorization URL so you can open it manually. You do not need to copy a credential from the browser back into the terminal.
API Key alternative
You can also skip browser authorization and provide an API key directly:
VOC_OPENAPI_API_KEY=voc_live_sk_xxx npx voc-openapi-mcp@latest setup
Keep the API key private: do not commit it, paste it into issues or chat, or include it in screenshots and logs. Prefer the environment variable above over putting the key directly in a shared command or script.
Preview
npx voc-openapi-mcp@latest setup --api-key voc_live_sk_xxx --dry-run
Note: To support a one-command setup, the API key is currently stored as plain text in the local MCP client configuration file.
Local stdio server for catalog scanners
Catalog scanners such as Glama primarily use the repository Dockerfile to build and start a minimal local stdio MCP server for evaluation. From a source clone or Docker image, the server command is:
node bin/voc-mcp.js serve
After the npm package release, the equivalent package command is:
npx voc-openapi-mcp@latest serve
This mode exposes the same seven curated tools through tools/list for
catalog introspection and Docker-based evaluation. Real tools/call requests
require VOC_OPENAPI_API_KEY, VOC_API_KEY, or X_API_KEY; without a key the
server returns an explicit API-key-required error instead of fake business
results. Production users should continue to use the hosted endpoint
https://open.voc.ai/mcp/catalog.
Example result
This single agent request combines product and listing data, review insights, customer pain points, competitive positioning, and recommended actions. Click the image to view the full result.
Capabilities
- Fetch current or historical Amazon reviews.
- Analyze sentiment, pain points, and themes.
- Generate listing improvements from customer feedback.
- Tag feedback and product listings into structured taxonomies.
See TOOLS.md for all seven tools, input fields, credit costs, and task-creating side effects.
Privacy, permissions, and support
The hosted server only receives the selected tool inputs and the X-API-Key
required for the authorized VOC account. The installer writes only the selected
AI client's MCP configuration and the optional VOC skill. Review
PRIVACY.md and SECURITY.md before installation.
License
The installer source code is licensed under the Apache License 2.0. The hosted VOC OpenAPI service and its business capabilities remain subject to the VOC Terms and Conditions.