TakeFlow · 鲲图OS

TakeFlow Local Connector · Setup Guide

Connector version v2.4.0 Generated 2026-11-19 macOS · Windows · Linux
Contents (10 sections)
  1. 1. What the Connector Does for You
  2. 2. Before You Begin
  3. 3. Installation
  4. 4. Everyday Usage Examples
  5. 5. Operations That Require Your Confirmation
  6. 6. Troubleshooting
  7. 7. Data and Privacy
  8. 8. Version, Upgrade, and Uninstall
  9. 9. Legal and Compliance Notice
  10. Appendix: Manual Configuration for IT

This guide covers installation, daily use, and troubleshooting for the TakeFlow local connector. Once Section 3 is complete, you can simply describe what you need in natural language — no commands or parameters to memorise.

A web version is available at 接入指南-EN.html in the same folder. It contains the same content and supports printing.

中文版:接入指南.md / 接入指南.html.

AudienceOperators using TakeFlow seats
Applies toLocal connector v2.4.0
Reading guideSections 2–3 for first-time installation; Sections 4–5 for daily use; Section 6 for troubleshooting

1. What the Connector Does for You

After installation, the AI assistant can operate your TakeFlow seats directly, within the permissions you grant:

The connector works with WorkBuddy, Codex, Claude Desktop, Cursor, DeepSeek Harness, and other desktop clients that can run local programs. Install once and it works across all of them. Cloud-only products such as ChatGPT cannot load a local connector; see Section 6 for the reason and the alternative.

Three notes on how it runs:

No server requiredThe connector runs on your own computer
No review waitUnzip, run the installer, and you are ready
Data stays localCredentials are stored on your machine; requests go directly from your computer to the platform

2. Before You Begin

1. A computer with internet access

macOS, Windows, and Linux are all supported.

2. Node.js 18 or later

Run node -v in a terminal (Command Prompt on Windows):

3. A TakeFlow account

This is the phone number and password you normally use to sign in to TakeFlow.

ffmpeg is optional and only needed for the "screen recording diagnosis" feature. If you only analyse screenshots, you can skip it. To install: brew install ffmpeg on macOS, or winget install Gyan.FFmpeg in PowerShell on Windows.

4. Python 3 (optional)

Only needed for the "ban check" feature. Run python3 --version; 3.8 or later is sufficient. macOS and most Linux distributions include it. On Windows, install from https://www.python.org/downloads/.


3. Installation

Step 1: Extract to a Fixed Folder

After extraction you will see two folders:

Suggested locations:

Two things to keep in mind:

  1. Do not run it from your Downloads folder; the system may clean that directory.
  2. Do not delete or move the folder after installation. The client configuration points to this path, and moving it requires reinstalling.

Step 2: Run the Installer

SystemHow to run
macOSDouble-click 一键安装.command. The system may block it on first run; see 3.1
WindowsDouble-click 一键安装.bat. If Windows SmartScreen appears, click "More info" → "Run anyway"
LinuxOpen a terminal in the folder and run bash 一键安装.sh

3.1 macOS Blocks the Installer on First Run

The installer is not signed with an Apple developer certificate, so macOS shows the following on first launch:

Apple could not verify "一键安装.command" is free of malware that may harm your Mac or compromise your privacy.

This is the standard warning for unsigned scripts, not a virus alert. Allow it once as follows:

  1. Click "OK" to dismiss the dialog.
  2. Open System Settings → Privacy & Security.
  3. In the Security section, you will see "已阻止「一键安装.command」以保护 Mac" with an "Open Anyway" button next to it.
  4. Click "Open Anyway" and authenticate with your password or Touch ID.
  5. Double-click 一键安装.command again and click "Open" in the dialog.

If "Open Anyway" does not appear, right-click the file, choose "Open", then check again. If it still will not run, open a terminal in the folder and run bash 一键安装.sh. This method is not affected by the double-click restriction.

3.2 The Installer Asks Four Questions

  1. API address — press Enter to use the default.
  2. Phone number — enter the phone number of your TakeFlow account and press Enter.
  3. Password — nothing is displayed while you type, including asterisks. This is expected. Press Enter when finished.
  4. Clients to configure — the installer asks about WorkBuddy, Codex, Claude Desktop, Cursor, and DeepSeek Harness in turn. Answer y or n for each. WorkBuddy, plus Codex and DeepSeek Harness when their directories are detected, default to yes; Claude Desktop and Cursor default to no.

When finished, you will see:

✓ Login successful, credentials saved
✓ Connection verified, 8 seats found in your account
✓ Installation complete!

"Installation complete" means you are done. The number of seats reflects your own account.

3.3 You Can Close the Terminal Afterwards

Yes. The terminal is only used to run the installer. The connector is not a background service; the client starts it on demand and it exits when the call finishes.

"Process completed" at the bottom of the window means the program exited normally.

Step 3: Restart Your Client and Verify

Quit the client completely and reopen it. For WorkBuddy, quit the entire application (Cmd+Q on macOS, or "Quit" from the tray icon on Windows); closing the window alone is not enough.

Then run a test request:

Show me my TakeFlow seats and highlight the ones expiring soon

The client may ask for permission to use the connector on the first call; allow it. If the seat list comes back, installation and connection are complete.


4. Everyday Usage Examples

Every scenario below can be handled in a single sentence. The AI assistant works out the steps; you do not need to know the underlying tools or parameters.

Listing Seats and Expiry Dates

What seats do I have, and which ones are expiring soon?

You will get seat numbers, bound accounts, follower counts, expiry dates, and power status. Seats nearing expiry are flagged separately.

Powering On, Powering Off, and Viewing the Screen

Power on seat 16 and send me the screen link when it is ready

The assistant confirms the target seat, powers it on, waits for the screen to become available, and returns a console link. Open the link and sign in once with your TakeFlow account to view the screen.

Note: seats are powered off when not in use. This does not affect scheduled tasks — the platform powers them on before a task starts and off after it finishes, with no manual intervention.

Renewals

How much does it cost to renew seat 29 for 3 months?

You will receive the plan and the amount payable (including fees). The order is placed only after you confirm, and a payment link is provided. You can then ask the assistant to check the order status.

Uploading and Publishing

Publish the file new-product.mp4 on my desktop to seats 52 and 53, reusing the previous caption

This takes two steps: the platform selects materials by folder and cannot target a single file, so the file must first be uploaded to a cloud folder, and the publishing task is then created from that folder.

You only need to confirm: which seats, how many items are in the folder, publish now or scheduled time, and the title and caption. The assistant handles everything else.

Rules worth knowing before publishing:

Updating Account Profiles

Replace the avatar of seat 16 with this image and change the nickname to your_nickname

The avatar is uploaded to cloud storage, its URL is obtained, and the profile is updated automatically. You do not need to handle formats yourself. Fields you do not mention are left unchanged: changing only the avatar keeps the nickname and bio as they are.

Querying Performance Data

How are the recent posts on seat 52 performing?

You will get an account overview (followers, likes, post count) plus data for the 10 most recent posts.

Two notes:

Analysing Failed Tasks

Why did yesterday's task fail to publish? (attach the recording path or link)

Please analyse this screenshot and tell me where it got stuck.

The assistant extracts key frames, compares them against the rule library, and returns four sections: analysis result, key frames, core judgement, and mitigation suggestions.

Checking Whether Accounts Are Banned

Please check whether these TikTok accounts still exist: @user123, @user456, @user789

Each account receives one of four verdicts — exists, not found, exists but all zeros, or unconfirmed — along with nickname, post count, and follower count. Ten accounts take about two minutes, which is normal.

Two notes:

Other Common Requests

Check the login status of all seats and sign in any that are offline

Create a daily activity task for seats 2 and 19, starting at 9:00 tomorrow morning

Add a customer record and enable CLAW


5. Operations That Require Your Confirmation

For the four categories below, the assistant restates the operation and waits for your confirmation before executing:

CategoryExamplesHow confirmation looks
Cost involvedRenewal orders"Renew seat 29 for 3 months, total ¥xxx including fees. Confirm the order?"
Outbound contentPublishing posts, changing avatar or nickname"Publish 1 item each to seats 52 and 53, tomorrow at 10:00, titled xxx. Confirm?"
State changesPower on/off, bulk sign-in, daily activity tasks"About to power on seat 16. Confirm?"
File uploadsUploading materials, creating folders"About to upload xxx.mp4 to the folder '20261026'. Confirm?"

In short: anything involving money, account state, or public-facing content is confirmed with you first.


6. Troubleshooting

Check the current version first (Section 8). Including it when you contact support will speed things up.

Double-clicking the Installer Does Nothing

Node.js Not Detected

Node.js is not installed, or the version is below 18. Install the LTS release from https://nodejs.org/en and run the installer again.

TakeFlow Does Not Appear in the Client After Installation

Check in this order:

  1. Did you fully quit and restart the client? Closing the window alone is not enough.
  2. In WorkBuddy, open Plugins → MCP Servers and check that the takeflow entry is healthy (green). Red indicates a configuration problem.
  3. In the installation folder, run node server/takeflow-mcp.mjs in a terminal. No output is normal; press Ctrl+C to exit. If an error appears, send it to support.

"Credentials Not Configured" or "Token Expired"

Run the installer again to sign in. Tokens expire every 30 days as a platform security measure; this is not a fault.

Login Fails with "Incorrect Account or Password"

Make sure you are using your TakeFlow platform account, not a TikTok account. If the phone number is not registered, the installer offers to sign up with the same password; answer y.

Image or Video Upload Fails

Screen Recording Diagnosis Reports Missing ffmpeg

Install ffmpeg as described in Section 2 and restart the client. Screenshot-only analysis does not require it.

Ban Check Cannot Find python3

This feature requires Python 3.8 or later. Install it as described in Section 2, item 4, then run the request again. If your computer does not allow installing Python, you can skip this feature; everything else still works.

Company Computers Do Not Allow Installing Node.js

Ask IT to use a portable Node build (unzip and run, no installation), and point the command field in the client configuration to that node binary. See the appendix.

Can ChatGPT Use This Connector Directly?

No. ChatGPT runs in the cloud and cannot launch programs on your computer. This connector runs locally, with credentials and operations staying on your machine, so it cannot be added to ChatGPT.

Using it inside ChatGPT would require a separate public gateway — domain, certificate, authentication, and ongoing operations. That is an independent service with its own cost, billed separately from the local package. WorkBuddy, Codex, Claude Desktop, Cursor, and DeepSeek Harness are desktop clients and are not affected.


7. Data and Privacy


8. Version, Upgrade, and Uninstall

Checking the Current Version

Any one of these three methods works:

  1. Check the archive name — for example TakeFlow-MCP-本地接入包-v2.4.0-20261119.zip. The number after v is the version; the last 8 digits are the build date (10 November 2026).
  2. Run the installer — the first line shows the version. You can close the window immediately; nothing is changed.
  3. Ask the AI assistant — "What version of the TakeFlow connector is installed, and how long until the token expires?" returns the details for your machine.

Including the version when you contact support will speed up diagnosis.

Component Versions

The package bundles one connector and six skills:

ComponentVersionPurpose
Local connectorv2.4.0Installer and MCP service
takeflow2.3.0Seats, power, renewals, tasks, materials, publishing, profiles, performance data
seats-check1.0.0Seat sign-in and account checks
cultivation-task1.0.0Daily activity tasks
client-info1.0.0Customer record entry
tiktok-auto-ops-diagnostic1.1.2Screen-recording diagnosis of failed tasks
tiktok-ban-check1.0.0Account ban checks

Skills ship with the package and are not upgraded separately.

Upgrading

Extract the new package over the existing folder (or into a new folder), run the installer again, then restart your client. Existing configuration and skills are backed up automatically. The local edition does not update itself; upgrades are manual.

Uninstalling

  1. Delete the ~/.takeflow directory (on macOS, press Cmd+Shift+G in Finder, enter ~/.takeflow, and press Return).
  2. Open the client's MCP configuration and remove the takeflow entry (DeepSeek Harness users should remove the takeflow-mcp entry from ~/.dsh/profiles/<profile>/cordis.patch.yml).
  3. WorkBuddy users should also delete six folders under ~/.workbuddy/skills/: takeflow, seats-check, cultivation-task, client-info, tiktok-auto-ops-diagnostic, and tiktok-ban-check. DeepSeek Harness users should delete the same folders under ~/.dsh/skills/.

This section explains the relationship between the connector and third-party platforms and trademarks, together with the limits of responsibility when you use it. Please read it before use.

Third-Party Names and Icons

The names, icons, interface screenshots, and trademarks of WorkBuddy, Codex, Claude, DeepSeek Harness, and other products that appear in this guide, the installation package, or the accompanying demonstration video belong to their respective owners. They are used only to describe the connector's compatibility and configuration locations. Their use does not imply any affiliation, partnership, agency, sponsorship, or endorsement, nor any certification or recommendation of this product by those companies. Such use does not challenge their rights; if a rights holder considers the usage inappropriate, we will adjust or remove it promptly upon notice.

MCP Is an Open, General-Purpose Protocol

The Model Context Protocol (MCP) is an open technical protocol and is not proprietary to this product. This connector is implemented according to that protocol and can therefore be called from MCP-capable clients. Evolution of the protocol itself, including version changes and compatibility, is determined by its specification maintainers. "MCP support" indicates technical compatibility only; it does not mean that any platform has certified, listed, or recommended this product.

Comply with Each Platform's Terms

When connecting to and using any AI platform, you must also comply with that platform's user agreement, developer terms, usage policies, and the laws and regulations of your jurisdiction. Each platform decides unilaterally — and may change, restrict, or discontinue at any time — the scope of MCP support, configuration method, available features, and rate limits. This product cannot guarantee continued availability on any platform and is not responsible for such platform decisions.

Scope of Use

This connector is intended solely for operating TakeFlow and TikTok accounts that you own or are authorised to manage. Do not use it to circumvent platform risk controls or review, to mass-register or trade accounts, to publish unlawful content, or for any other activity that breaches platform rules or applicable law. You are responsible for the operations performed under your accounts and their consequences.

Third-Party Data and Services

Playback statistics, ban-check results, and similar outputs come from third-party public data sources and may be delayed, truncated, or incomplete. They are provided for operational reference only and do not constitute an official platform conclusion or legal evidence.

Limitation of Responsibility

This guide describes technical integration only and does not grant any authorisation, agency, or warranty with respect to any third-party platform. Where client version changes, platform rule changes, network conditions, or third-party data source failures cause integration failures, limited functionality, or data discrepancies, we will assist with diagnosis but are not liable for indirect losses arising from them.


Appendix: Manual Configuration for IT

This section is for IT staff. Regular users only need to run the installer as described in Section 3.

A1. Obtaining a Token Manually

curl -X POST https://saas.takeflows.com/admin-api/system/tk-auth/login \
  -H 'Content-Type: application/json' \
  -d '{"phoneNumber":"13800000000","password":"your-password"}'

Write the returned accessToken and refreshToken into ~/.takeflow/config.json and set its permissions to 600:

{
  "baseUrl": "https://saas.takeflows.com",
  "phoneNumber": "13800000000",
  "accessToken": "the returned accessToken",
  "refreshToken": "the returned refreshToken",
  "expiresTime": 1792633796480
}

A2. Client Configuration

ClientConfiguration file
WorkBuddy~/.workbuddy/mcp.json (Windows: %USERPROFILE%\.workbuddy\mcp.json)
Codex CLI~/.codex/config.toml
Claude DesktopmacOS ~/Library/Application Support/Claude/claude_desktop_config.json; Windows %APPDATA%\Claude\claude_desktop_config.json
Cursor~/.cursor/mcp.json
DeepSeek Harness~/.dsh/profiles/<profile>/cordis.patch.yml (the desktop profile is desktop)
{
  "mcpServers": {
    "takeflow": {
      "command": "/absolute/path/to/node",
      "args": ["/absolute/path/to/TakeFlow-MCP/server/takeflow-mcp.mjs"],
      "env": { "TAKEFLOW_HOME": "/Users/your-name/.takeflow" }
    }
  }
}
[mcp_servers.takeflow]
command = "/absolute/path/to/node"
args = ["/absolute/path/to/TakeFlow-MCP/server/takeflow-mcp.mjs"]

[mcp_servers.takeflow.env]
TAKEFLOW_HOME = "/Users/your-name/.takeflow"
# DeepSeek Harness: written to cordis.patch.yml inside the profile directory
- id: takeflow-mcp
  name: '@deepseek-ai/dsh-mcp-client'
  config:
    serverName: takeflow
    transport: stdio
    command: "/absolute/path/to/node"
    args:
      - "/absolute/path/to/TakeFlow-MCP/server/takeflow-mcp.mjs"
    env:
      TAKEFLOW_HOME: "/Users/your-name/.takeflow"
    toolCallTimeoutMs: 300000
    failOnStartupError: true

Three extra points for DeepSeek Harness: edit cordis.patch.yml, not the cordis.yml beside it; fully quit and relaunch the app after installing, since the plugin tree is composed only at startup; and keep toolCallTimeoutMs above the 60-second default, otherwise video frame extraction will be cut off while the clip downloads.

Two notes:

A3. Codex Skill Index

Codex has no skill mechanism, so the installer maintains an index block in ~/.codex/AGENTS.md between <!-- takeflow-skills:start --> and the matching end marker. It tells the model which document to read in which situation. Content outside the block is left untouched.

For detailed fields and parameters, see README.md in the same folder.


If you run into problems during installation or use, send a screenshot of the terminal error to support, or simply ask the AI assistant: "I got the following error while installing the TakeFlow connector — how do I fix it?"

Generated from 接入指南-EN.md. Edit the Markdown and regenerate; do not edit this file directly.