Table of Contents
Overview
The Trovata MCP Connector gives any MCP-compatible AI assistant direct access to your Trovata data. Ask a question in plain English, and the assistant retrieves your balances, transactions, tags, and reports in seconds.
MCP stands for Model Context Protocol, an open standard originally developed by Anthropic and now adopted across the industry, including by Google, Microsoft, and OpenAI. It lets AI assistants connect to the tools and data you already use, without a custom integration for every platform.
Getting Started
Before you Start: You'll need an active Trovata account with your banks connected, and an MCP-compatible AI client (e.g. Claude Desktop, Claude Code, Cursor, or any client that supports remote MCP servers).
Trovata's MCP server connects directly to the AI assistant you already use. Instead of company-level credentials, each person signs in with their own Trovata login, so responses are scoped to that person's own entitlements automatically.
Server URL: https://user.mcp.trovata.io
OAuth client ID: ATROlFfEPkYAJ0xd8pkYOFL0tfnWwMvF (This is a public ID and safe to share)
Setup differs slightly by assistant. Find your tool below.
Claude
Works with claude.ai on the web and the Claude desktop app.
Claude Code and the Claude CLI
Run one command, then authenticate:
bash
claude mcp add --transport http trovata https://user.mcp.trovata.io/mcp
Start claude, run /mcp, select trovata, then Authenticate. Sign in with your Trovata credentials in the browser window that opens. There's no client ID or port configuration to enter.
Claude Web and Desktop
Open Settings, then Connectors, then Add custom connector.
Enter the URL: https://user.mcp.trovata.io
Expand Advanced settings and enter the OAuth client ID above. Leave the client secret blank.
Select Add, then Connect. Sign in with your Trovata credentials when the login window opens.
In any chat, open the tools menu, enable the Trovata connector, and ask a question, for example: "What's my total cash position across all banks?"
To switch companies later, disconnect and reconnect the connector, then sign in again.
Gemini CLI
Gemini's consumer app doesn't accept custom connectors, but the Gemini CLI does.
Open ~/.gemini/settings.json (create it if it doesn't exist).
Add the Trovata server to mcpServers:
JSON
{
"mcpServers": {
"trovata": {
"httpUrl": "https://user.mcp.trovata.io/mcp",
"oauth": {
"enabled": true,
"clientId": "ATROlFfEPkYAJ0xd8pkYOFL0tfnWwMvF",
"authorizationUrl": "https://login.trovata.io/authorize?audience=https%3A%2F%2Fuser.mcp.trovata.io%2F",
"tokenUrl": "https://login.trovata.io/oauth/token",
"scopes": ["openid", "profile", "email"],
"audiences": ["https://user.mcp.trovata.io"]
}
}
}
}
Start gemini, run /mcp auth trovata, and sign in with your Trovata credentials in the browser window that opens.
Confirm the connection with /mcp. The Trovata tools appear once authentication completes.
Gemini Enterprise
Workspace admins can register Trovata as a custom connector for the whole organization under Admin console, then Data stores, then Custom MCP server, using the server URL above with the StreamableHTTP transport. Contact your Trovata representative to coordinate the OAuth client registration for your tenant.
GitHub Copilot
Works in VS Code 1.101 or later, and in the same server works in Visual Studio, JetBrains, and other Copilot IDEs that support remote MCP.
Open the Command Palette and run MCP: Add Server, then choose HTTP.
Enter the URL: https://user.mcp.trovata.io/mcp
Name it trovata and save to your user configuration. VS Code writes this to mcp.json:
Select Auth when VS Code prompts for authentication. When asked for a client ID, enter the OAuth client ID above and leave the secret blank, then sign in with your Trovata credentials.
Open Copilot Chat in agent mode. The Trovata tools appear in the tools picker.
ChatGPT
ChatGPT custom connectors run through developer mode, available on Pro, Plus, Team, Enterprise, and Edu plans.
Open Settings, then Connectors, then Advanced settings, and enable Developer mode.
Select Create connector and enter the URL:
https://user.mcp.trovata.ioSet Authentication to OAuth.
Under OAuth advanced settings, set Registration method to Client Identifier Metadata Document (CIMD).
Select Create, then sign in with your Trovata credentials and approve the consent screen.
Note: ChatGPT support requires automatic client registration on Trovata's side. If the connection is declined at sign-in, contact your Trovata CSM to have it enabled for your organization.
OpenAI Codex
Add the server, then authenticate from the app. Codex registers itself and captures the login on a local callback, so there's nothing to configure. The browser returns you to a local address when the login completes — that's expected.
Sample Prompts
The MCP Connector exposes the same capabilities available through Trovata's Developer API, delivered through conversation instead of code.
Cash visibility
"What's our consolidated cash position right now?"
"Show me balances across all accounts at JPMorgan Chase."
"What's the historical balance trend for our operating accounts over the past 90 days?"
"How does our cash position today compare to this time last month?"
Transaction intelligence
"Find all transactions over $50,000 in the past 30 days."
"Show me a summary of credit versus debit activity this month by currency."
"What were our largest outflows last week? Break them down by tag."
"Look up transaction ID [xyz] and give me the full details."
Tag-based analysis
"Show me all transactions tagged as 'Intercompany' this quarter."
"What's our net cash flow for 'Payroll' tagged transactions by entity?"
"List all active transaction tags and how many transactions each contains."
Reporting
"What reports do I have available?"
"Run the monthly cash overview report for March."
"Compare this month's cash report against last month's."
Invoice Management
"Show me all outstanding invoices."
"What's the total invoiced amount by company this quarter?"
Multi-Dimensional Analysis
This is where the MCP Connector adds the most value beyond the raw API. Multi-dimensional queries that would normally require nested parameters and manual correlation become a single question.
"Break down our cash inflows by currency and region for Q1."
"What's our monthly debit trend by entity over the past 6 months?"
"Compare transaction volume across our top 5 banks this quarter."
Best Practices
Be specific about time. "Last 30 days" or "Q1 2026" gives the assistant a clear window. "Recently" leaves too much room for interpretation.
Use your account nicknames. This gives the assistant something concrete to search for.
Use follow-up questions. Most MCP-compatible assistants maintain context within a session, so you can build on a previous answer without re-explaining it.
Security and Governance
Summary: The MCP Connector authenticates through your own Trovata login via OAuth, not a shared company-level credential. Every response reflects your individual entitlements, not admin-level access.
Authentication. You sign in with your own Trovata credentials through OAuth. The only shared value is a public OAuth client ID, which isn't sensitive and doesn't grant access on its own. There's no App Secret or company-level credential to distribute or protect.
Authorization. Every request runs under the signed-in user's own role-based entitlements in Trovata. A response reflects exactly what that person is permitted to see, not admin-level access, and not a company-wide credential shared across users.
Data residency. Your treasury data is served directly from Trovata's infrastructure. It passes through the AI assistant's context for the duration of the conversation only and is not stored by the AI provider.
Encryption. All MCP connections use TLS 1.3 encryption in transit. Data at rest remains encrypted within Trovata's SOC 2-certified infrastructure.
Note: Because access is scoped to each individual's own Trovata login, there's no shared credential to manage or restrict distribution of. If someone needs different access, that's controlled the same way it already is in Trovata: through their role-based entitlements, not through who holds a connector credential.
MCP Connector vs. Developer Portal APIs
Trovata's Developer API and the Trovata MCP Connector access the same underlying treasury data. The difference is who each one is built for.
The MCP Connector is a conversation layer for everyone else. There are no API keys to manage and no responses to parse. You ask a question, and the AI assistant handles the rest.
The Developer API is a building block for engineering teams. It requires a developer who understands authentication, HTTP requests, and JSON parsing. It's the right tool for data pipelines, ERP syncs, and custom dashboards in PowerBI, Tableau, etc.
Both access the same Trovata data. Neither replaces the other.
Dimension | MCP Connector | Developer API |
Intended User | Treasury analysts, controllers, CFOs, anyone with Trovata access | Engineers and developers |
Setup | One-time connection in your AI assistant | API keys, custom code, ongoing maintenance |
How you Interact | Natural language questions | HTTP requests with structured parameters |
Output | Conversational answers, tables, and summaries | Raw JSON |
Maintenance | Trovata maintains the connector | Your team maintains the integration |
Time to First Value | Minutes | Days to weeks |
FAQs & Troubleshooting
Does the MCP Connector replace the Developer API? No. The Developer API remains available for engineering teams that need programmatic access for data pipelines, ERP integrations, and custom applications. The MCP Connector is an additional access layer optimized for AI-powered interaction. Both access the same underlying data.
Do I need to install anything? No. The MCP Connector is hosted by Trovata. You connect to it from your AI assistant using a server URL, and you sign in with your own Trovata login. There's no local installation and no SDK.
Which AI assistants are supported? Claude (web and desktop), the Gemini CLI, GitHub Copilot (in VS Code, Visual Studio, and JetBrains IDEs that support remote MCP), ChatGPT (developer mode, on Pro, Plus, Team, Enterprise, and Edu plans), and Gemini Enterprise for Workspace admins. Note that Gemini's consumer app doesn't currently accept custom connectors, so Gemini access runs through the CLI or Gemini Enterprise instead. The list continues to grow as more AI tools adopt the MCP standard.
Can the AI modify my data or initiate payments? No. The MCP Connector provides read-only access to your treasury data, the same data available through the Developer API today.
Is there additional cost for MCP access? During the AI 2.0 trial, the MCP Connector is free to use. After the trial, it will be a paid premium feature.
How is my data protected? You connect using your own Trovata login through OAuth, so there's no shared App Secret or company-level credential to manage, and every response reflects your own permissions and entitlements rather than a company-wide access level. The connection runs over TLS-encrypted transport within Trovata's SOC 2-certified infrastructure. The AI assistant receives data only within the context of your conversation session and does not persistently store your treasury data.
What if I have questions or need help getting connected? Reach out to your Trovata Customer Success Manager, or contact support@trovata.io. We'll walk you through the connection process and help your team get the most from AI-powered treasury intelligence.
Troubleshooting "Couldn't connect" in Claude Remove the connector and re-add it. Claude caches connector settings from the first attempt.
Troubleshooting Login Loops or "service not found" Confirm the URL ends with a trailing slash in the connector settings: https://user.mcp.trovata.io/
Troubleshooting A tool Returning "your Trovata permissions don't include this" Your entitlements don't cover that data. Ask your Trovata admin if you believe you should have access.