Skip to main content

Connecting to Flourish with an LLM using our MCP Server

Connect Flourish to Claude and other LLMs via MCP server to query inventory, update attributes, and manage package locks. Read tools always allowed; write actions require approval.

Use Flourish's MCP server to access tools from your LLM (Large Language Model) interface of choice, including Claude, ChatGPT, Gemini, Grok, etc.

What's an MCP Server? The Model Context Protocol (MCP) is a new way for AI tools to communicate with software applications. The Flourish MCP server exposes a set of tools to perform certain actions in Flourish using a LLM chat interface.

What tools are available? We're working closely with accounts to safely expose functionality through this interface, and we will regularly update the tool list as we add new capabilities.

  • Querying and updating Item Attributes. Note: The initial release only supports updating attributes that do not sync with a compliance system.
  • Querying package data
  • Locking and unlocking packages

Connect Flourish Software to Claude (claude.ai chat)

You'll need a Claude.ai Pro/Max/Team/Enterprise subscription (custom connectors aren't available on Free).

Follow these steps to connect your Flourish account to Claude using Flourish's MCP server. Claude only sees data you have permission to access in Flourish, scoped to your user's facility access. You can opt out of exposing data to AI training when you configure your Claude account.

First, click "Customize" to show the customization screen. Navigate to Connectors, click the plus sign, and click "Add customer connector."

Then, name the connect. For example, "Flourish Software." Enter the Flourish MCP Server URL into the screen, and click "add."

Flourish Server URL: https://flourish-mcp-server.prd.k8s.flourishsoftware.com/mcp

Click "Connect" to open the login page.

Sign in with your normal Flourish username and password. Client name is used to identify the tool, so you can put "Claude" or "ChatGPT" here.

Modify the tool permissions to align with your comfort level for tool approval.

Note: We will be updating the tools on a regular basis, so you may need to refresh the tool list to access new tools.

Claude lets you choose how each tool is gated. We recommend the following starting configuration. You can tighten or loosen these later as you build trust with the workflow.

Rule of thumb: any tool name ending in _query is read-only and safe to allow freely. Anything that writes to your data and should stay on "Ask every time" until you're comfortable with how Claude is using them. We will never auto-elevate a read tool to a write tool. New write tools always arrive on the "Needs approval" default.

Tool

Type

Recommended Setting

Notes

user_whoami

Read

Always allow

Identifies your session. No data exposure.

user_hello

Read

Always allow

Diagnostic ping.

user_facilities

Read

Always allow

Lists facilities you already have access to.

tool_search

Read

Always allow

Discovery only; doesn't return business data.

uoms_query

Read

Always allow

Static reference data.

lock_codes_query

Read

Always allow

Reference data for the lock workflow.

items_query

Read

Always allow (or Ask)

Returns inventory item data. Set to Ask if you handle sensitive catalogs.

packages_query

Read

Always allow (or Ask)

Returns package-level inventory. Set to Ask if you handle sensitive batch data.

integration_sync_statuses_query

Read

Always allow

Read-only health check on Metrc/BioTrack syncs.

items_update

Write

Ask every time

Mutates item records. Always review the proposed change before approving.

packages_lock

Write

Ask every time

Changes package state. Review the package list and lock code before approving.

packages_unlock

Write

Ask every time

Clears lock state on packages. Review the package list before approving.

Example Prompts

Once connected, try these to get a feel for what the integration can do. Claude will ask you to choose a facility on the first prompt of a session if you have access to more than one.

Inventory & operations

  • "What facilities do I have access to in Flourish?"
  • "Show me all packages at [facility name] that are currently locked, and why."
  • "List items where the SKU starts with 'GUM-' and the product class is edible."
  • "How many packages of [strain name] do I have on hand, grouped by facility?"

Compliance & sync health

  • "What's the current sync status of my Metrc integration? Are there any failed records?"
  • "Show me any integration sync errors from the last 24 hours and explain what they likely mean."

Lock workflow (write actions — Claude will ask before executing)

  • "Find all packages at [facility] tagged with the test batch ID 1A4… and lock them with the 'Pending QA Review' lock code."
  • "Unlock the three packages I just locked for QA — testing came back clean."

Reference & exploration

  • "What units of measure does Flourish support, and which ones are 'count' vs 'weight'?"
  • "What tools does the Flourish MCP server expose? Group them by read vs write."

Tip: Claude works best when you give it a facility name or ID up front and tell it whether you want a summary, a table, or raw data. The more specific the prompt, the better the result.

Troubleshooting

The "Connect" button does nothing, or the login window closes immediately. Pop-ups may be blocked. Allow pop-ups for claude.ai in your browser and try again. If you're using a corporate browser profile, your IT policy may also be blocking third-party OAuth windows.

The login screen shows a countdown timer and times out before I finish. The sign-in window is short-lived (about 15 minutes) for security. If it expires, close the window and click "Connect" again. A fresh login window will be issued. Have your password and 2FA device ready before you start.

"Invalid credentials" even though my Flourish login works in the app. Make sure you're using the email that you log into Flourish with.

My password expired or was reset after I connected. Disconnect the connector (three-dot menu → Disconnect), then click Connect again and sign in with your new password. Existing chats won't retroactively reauthenticate.

Tools are missing, or new tools you announced aren't showing up. Open the connector settings and click "Refresh tools list" at the top right. We add and update tools regularly; refreshing pulls the current catalog without requiring you to disconnect and reconnect.

Claude says it "doesn't have access" to a facility I should be able to see. The MCP session inherits the same facility access as your Flourish user account. If a facility is missing, check your user's facility permissions inside Flourish first. If your access is correct there, disconnect and reconnect the connector to refresh your session token.

A write action (lock, unlock, update) failed. Check the error message Claude returns. Common causes include: the package is allocated to an open order or manufacturing run (which blocks locking), an invalid lock_code_id for that facility (lock codes are facility-specific), or insufficient role permissions on your Flourish user. Re-running the read query first (e.g., packages_query) will usually surface the underlying state.

How do I disconnect or revoke access? In Claude, open the connector and choose "Disconnect" from the three-dot menu. This invalidates the session immediately. To audit or revoke access from the Flourish side, contact support — full self-service session management is on the roadmap.

Something else is broken. Email support with: your Flourish username, the approximate time of the issue, the prompt you sent Claude, and any error message Claude displayed. Screenshots help.

How did we do?

CYBRA MarkMagic® Integration Setup

Contact