MCP Setup (Claude, Cursor, Codex)

MCP setup is the compounding step, not the product by itself. The fastest path is still: pick a project, define rules, capture context, generate a pack, then wire your assistants into that setup so repeated instructions disappear from future sessions.

1. Pick a project
2. Define rules
3. Save shared context
4. Generate a pack
5. Add veriova MCP so every future session starts with the same context and rules

For Claude Code, Cursor, and Codex-run setup from your project directory:

npx veriova-mcp@latest setup

At the API key prompt, type open to open signup, generate an API key, then paste it back into the terminal.

Setup configures your MCP client to launch npx veriova-mcp with environment variables. It does not upload your API key anywhere except to veriova API calls you explicitly make.

VERIOVA_API_KEY=cv_...
VERIOVA_BASE_URL=https://veriova.com
VERIOVA_PROJECT=<selected project slug>

Add veriova to your Claude Desktop config file:

{
  "mcpServers": {
    "veriova": {
      "command": "npx",
      "args": ["veriova-mcp@latest"],
      "env": {
        "VERIOVA_API_KEY": "cv_your_key_here",
        "VERIOVA_PROJECT": "your-project-slug",
        "VERIOVA_BASE_URL": "https://veriova.com"
      }
    }
  }
}

On Windows, the config file is at %APPDATA%\Claude\claude_desktop_config.json. Restart Claude Desktop after editing.

Claude Chat supports remote MCP servers. No local setup needed. Connect via:

1. Open Claude Chat and go to Settings → Integrations → Add MCP Server.
2. Enter the remote MCP URL: https://mcp.veriova.com/mcp
3. You'll be redirected to sign in with your veriova account. The OAuth flow creates an API key automatically, so there's no manual key pasting.
4. After authorization, tools like veriova_session_start, veriova_readiness_review, veriova_check_standards, veriova_remember, veriova_recall, veriova_context, and veriova_list will appear in your chat.

Add this to your project instructions or system prompt so your assistant automatically loads team context and can run optional reviews after substantial code changes.

At the start of every conversation, call veriova_session_start to load your team's decisions, conventions, and production standards. Use this context throughout the conversation.

At the end of any session where you built or significantly modified code, call veriova_readiness_review. Write a technical description of what you built: what it does, how it handles authentication, errors, logging, retries, and any security concerns you addressed or deferred.

The current remote MCP surface is designed around four jobs: load session context, inspect rules, run optional reviews, and store or retrieve durable context.

• veriova_session_start loads pinned decisions, verified memory, and team rules at the start of a session.
• veriova_check_standards lists the team's criteria before building or reviewing.
• veriova_readiness_review scores a feature against those rules and returns gaps plus a shareable card URL.
• veriova_remember, veriova_recall, veriova_context, and veriova_list handle durable context storage and retrieval.

During setup, veriova fetches your projects and lets you choose from a list. The default suggestion is derived from your current folder name (workspace slug).

Example: if you run setup inside a folder named veriova, it will suggest veriova as a slug.

# Claude Code
claude mcp list

# Codex
codex mcp list

If tools are enabled, your assistant should be able to call:veriova_session_start, veriova_readiness_review, veriova_check_standards, veriova_remember, veriova_recall, veriova_context, veriova_list.

npx veriova-mcp@latest doctor

This checks connectivity, API key validity, and whether the configured project slug exists.

On some Cursor builds, the CLI MCP add command is unreliable. Setup automatically falls back to writing .cursor/mcp.json in your repo and adds .cursor/ to .gitignore to avoid committing secrets.