TFlow MCP Integration Guide (Connecting External AI Chatbots)
This guide explains how to connect external AI chatbots such as Claude Desktop and ChatGPT to TFlow. It walks through the available tools, the setup for each platform, usage scenarios, the built-in prompts, and frequently asked questions.
Last updated: 2026-04-01
Overview
With TFlow MCP (Model Context Protocol), you can access TFlow data directly from external AI chatbots such as Claude Desktop and ChatGPT.
You can check project status, look up documents, manage tasks, and request IRR analysis conversationally — handling your work without needing to log in to the web separately.
Supported Platforms
| Platform | Connection Method | Status |
|---|---|---|
| Claude Desktop | MCP Server (stdio) | ✅ Supported |
| ChatGPT | MCP Streamable HTTP + OAuth 2.1 | ✅ Supported |
| Other MCP-compatible apps | MCP protocol-compatible clients | ✅ Supported |
What Can You Do?
| Category | Example Questions |
|---|---|
| Project status | "Show me my project list", "What's the status of project 143?" |
| Task management | "This week's to-dos", "Mark the task as done", "Add a new task" |
| Document management | "Check the uploaded documents", "L/C document analysis results", "Upload a document" |
| Notifications | "Any new notifications?", "What's urgent?", "Mark notifications as read" |
| Customs analysis | "Run an IRR analysis", "Are we ready for customs?" |
| L/C risk | "Show me the LEO report" |
| Briefing | "What do I need to do today?", "Morning briefing" |
| Comments | "Leave a note on the project" |
| Token/system | "Check my token balance", "Show me the system notices" |
Available Tools
TFlow MCP provides the following tools:
| Category | Tool | Description |
|---|---|---|
| Projects | get_projects | Look up the project list and status |
get_project_overview | Detailed project status (trade info, documents, members) | |
| Tasks | list_my_tasks | Look up the to-do/task list |
create_task | Create a new task | |
update_task | Change task status, edit contents | |
| Documents | get_documents | Look up the project document list |
get_document_detail | Document details (including AI analysis results) | |
upload_document | Upload a trade document | |
| Notifications | get_notifications | Look up the notification list |
get_urgent_alerts | Consolidated view of urgent alerts | |
get_daily_briefing | Today's consolidated work briefing | |
mark_notifications_read | Mark notifications as read | |
| Analysis | get_irr_report | Look up the IRR (Import Readiness) report |
request_irr_analysis | Request IRR analysis (asynchronous, 1-3 min) | |
get_leo_report | Look up the LEO L/C risk analysis report | |
| Comments | add_project_comment | Add a comment to a project |
| System | get_token_balance | Look up AI token balance and usage |
get_system_context | User profile, system status | |
get_server_notices | System notices | |
get_recent_activity | Recent activity feed |
Setup
There are two ways to connect. Claude Desktop connects with an API key (PAT) issued in TFlow and written into a config file, while ChatGPT connects via OAuth login (Google/Kakao). Pick the one that matches the app you already use.
Method 1: Connect from Claude Desktop
Step 1: Issue an API Key in TFlow
- Log in to TFlow on the web
- Go to Settings → My Subscription → External Integrations tab
- Click Issue API Key
- Copy the issued key (⚠️ it is shown only once!)
💡 The API key looks like
pat_tflow_.... Keep it somewhere safe.
Step 2: Connect TFlow to Claude Desktop
- Launch the Claude Desktop app
- Go to the Settings → MCP Servers menu
- Add the TFlow server:
Option A: Edit the config file directly
Open the ~/Library/Application Support/Claude/claude_desktop_config.json file:
{
"mcpServers": {
"tflow": {
"command": "node",
"args": ["/path/to/tflow-mcp-server/index.js"],
"env": {
"TFLOW_API_URL": "https://api.tflowx.com",
"TFLOW_API_KEY": "pat_tflow_여기에_발급받은_키_붙여넣기"
}
}
}
}
Option B: Copy from the TFlow web
Clicking the Copy Claude Desktop config button on the External Integrations tab automatically copies the config above to your clipboard.
- Restart Claude Desktop
Step 3: Verify the Connection
In Claude Desktop, ask something like:
"Show me my TFlow project list"
If the project list appears, the connection was successful!
Method 2: Connect from ChatGPT
ChatGPT connects to TFlow using OAuth 2.1-based authentication. Without having to manage an API key directly, you can authenticate easily with social login (Google/Kakao).
Step 1: Find TFlow in ChatGPT
- Open ChatGPT and check the Tools or Plugins menu at the bottom of the message input box.
- Search for TFlow in the MCP server list.
Step 2: OAuth Login
- When you select the TFlow connection, the TFlow login screen pops up.
- Log in with your Google or Kakao account (it must be the same account registered with TFlow).
- On the authorization screen, allow ChatGPT to access your TFlow data.
Step 3: Verify the Connection
"Show me my TFlow project status"
If the project list appears, the connection was successful!
💡 Claude Desktop vs ChatGPT differences:
- Claude Desktop: PAT (API key) based — enter the key directly in the config file
- ChatGPT: OAuth 2.1 based — automatic authentication via social login, with automatic token refresh
Usage Scenarios
Starting Your Workday
"Give me a morning briefing"
→ Check today's to-dos, upcoming deadlines, and urgent alerts all at once
"Check my unread notifications"
→ Review completed document analyses, comments, and more
Getting a Grasp of Project Status
"Show me the status of project 143"
→ Trade info, document status, task progress, member list
"List this project's documents"
→ Check uploaded documents and their analysis status
"Show me the CI document analysis results"
→ Review the detailed data extracted by AI
Customs Readiness Analysis (IRR)
"Run a customs analysis on project 143"
→ Request IRR analysis (takes 1-3 min)
"Check notifications" (a moment later)
→ See the "IRR analysis complete" notification
"Show me the IRR results"
→ Detailed report of Blocker, Warning, and needs-review items
Task Management
"To-dos due this week"
→ List of tasks with upcoming deadlines
"Mark the Health Certificate issuance request task as done"
→ Change the task status
"Add a 'review shipping documents' task to project 143"
→ Create a new task
L/C Risk Analysis
"Show me the LEO report for project 143"
→ Review the Letter of Credit risk analysis results
"Pull out just the HIGH risk items"
→ AI extracts the high-risk items from the report
Built-in Prompts
You can use TFlow prompts directly in Claude Desktop:
| Prompt | Description |
|---|---|
tflow_start_here | Get started with TFlow — profile, feature guide, notice check |
tflow_morning_briefing | Morning briefing — urgent alerts + today's to-dos + weekly status |
tflow_project_review | Project review — detailed status of a specific project |
tflow_whats_new | What's new — system notices + recent changes |
💡 In ChatGPT, you can freely ask questions similar to the prompts above. You don't need to use the prompt names — just ask in natural language.
FAQ
I lost my API key
Go to Settings → My Subscription → External Integrations and click the Reissue button. The existing key is discarded immediately and a new key is issued.
I get an "API key is invalid" error in Claude
- The API key may have expired (default 90 days). Reissue it.
- Check that the key in the config file was copied exactly (watch for leading/trailing whitespace).
- Try restarting Claude Desktop.
TFlow won't connect in ChatGPT
- Check that you logged in with the same Google/Kakao account you registered with TFlow.
- If pop-ups are blocked, allow them in your browser settings.
- If the OAuth authentication token has expired, disconnect and reconnect TFlow in ChatGPT.
I can't see a specific project's data
You can only access projects you are a member of. Check the Parties tab of that project on the TFlow web.
Can I view document contents?
You can look up document metadata (file name, analysis status, etc.) and AI analysis results. To download the original file, you can get an S3 URL via the view: "full" option of get_document_detail.
Are tokens deducted?
Data lookups through MCP (projects, documents, notifications, etc.) do not consume tokens. However, IRR analysis (request_irr_analysis) runs an AI agent, so tokens are deducted.
Which is better, Claude Desktop or ChatGPT?
Both platforms access the same TFlow tools. Just use the AI chatbot you normally use. Claude Desktop is simple to set up with the API key method, while ChatGPT connects more conveniently via social login.