FAQ

FAQ

General

What is MCP?

Model Context Protocol is an open standard that lets AI assistants connect to external tools and data sources. Think of it as a USB port for AI — plug in any tool and the AI can use it.

Which AI assistants work with this?

Any MCP-compatible assistant:

• Claude Code (CLI)
• Claude Desktop
• Any app using the MCP SDK (Python, TypeScript, Ruby, Go)

Is this safe to use?

• The server runs locally on your machine — no data goes through third-party servers
• Credentials are stored in your local .env file
• Order placement always requires AI confirmation before execution
• The code is fully open-source — audit it yourself

Does this cost money?

The MCP server itself is free and open-source. However:

• Free tools (25): portfolio, orders, smart orders, instruments, profile — work with any Groww account
• Paid tools (5): quote, LTP, OHLC, option chain, historical data, backtesting — require Groww's Live Data subscription at ₹499 + GST/month

---

Authentication

Which auth method should I use?

TOTP — always. One-time 2-minute setup, then it works forever with zero daily maintenance.

My token expired. What do I do?

• TOTP auth: Just restart the server. It auto-generates a fresh token.
• Approval auth: Visit Groww → Trading APIs → click "Approve", then restart.
• Manual token: Generate a new token on the Groww website.

I get "Invalid credentials" with TOTP

Regenerate the TOTP token on the Groww website. The TOTP secret might have changed.

Can I use multiple auth methods at once?

Yes, but the server picks the first available in priority order: TOTP > Approval > Manual.

---

Tools

Why do market data tools return 403?

They require the Groww Live Data subscription (₹499 + GST/month). Subscribe at: Groww App → Settings → Trading APIs.

Can this place real orders?

Yes. place_order, modify_order, and cancel_order execute real trades with real money. The AI assistant will always confirm with you before placing an order.

What are GTT/OCO orders?

• GTT (Good Till Triggered): Sets a price trigger — when the stock hits your target, an order is automatically placed
• OCO (One Cancels Other): Sets both stop-loss AND target — whichever hits first executes, the other cancels automatically

Why is this the only Groww MCP with GTT/OCO?

Other implementations don't cover the /v1/smart-order/* endpoints. We verified and implemented every available Groww API endpoint.

---

Technical

What Ruby version do I need?

Ruby >= 3.1.0. Tested on Ruby 4.0.1.

Can I use this with rbenv/rvm/asdf?

Yes. Specify the full Ruby path in your .mcp.json config. See the Getting Started page for RVM-specific setup.

Why Ruby and not TypeScript/Python?

• The official MCP Ruby SDK is well-maintained (Anthropic + Shopify)
• Ruby's expressiveness makes tool definitions clean
• Strong HTTP/JSON support in stdlib
• Existing TypeScript and Python implementations had broken API paths

How does auto-refresh work?

The server reads the token's expiry from the JWT payload and refreshes 5 minutes before expiry. For TOTP auth, it regenerates a fresh code and exchanges it for a new token. This happens silently — no user action needed.

Why are other Groww MCPs broken?

Most use incorrect API endpoint paths (e.g., /portfolio/holdings instead of /v1/holdings/user). We verified every endpoint against the live Groww API and the official Groww Trade API documentation.

---

Troubleshooting

Server won't start

1. Check Ruby version: ruby --version (needs >= 3.1.0)
2. Run bundle install to ensure gems are installed
3. Verify .env has valid credentials

Orders fail with "static IP required"

Groww requires a registered static IP for order placement (SEBI compliance). Register yours at: Groww → Settings → Trading APIs → Add Static IP.

Server starts but AI can't connect

1. Verify the path in .mcp.json is correct
2. If using RVM, set GEM_HOME and GEM_PATH in the env config
3. Restart your AI assistant after changing .mcp.json

I see "rate limited" errors

Groww has strict rate limits (10 req/sec for orders, 20 req/sec for reads). The server auto-retries with backoff, but avoid rapid-fire queries.