Skip to content
Tallyfy
Developers

Using Tallyfy MCP server with ChatGPT

Run your Tallyfy work from ChatGPT

Section titled “Run your Tallyfy work from ChatGPT”

Connect Tallyfy to ChatGPT and you can manage your workflows just by typing what you need. Ask ChatGPT to find tasks, review a template, or pull a report, and it does the work in your Tallyfy account for you. Full access works on ChatGPT Business, Enterprise, and Edu plans.

ChatGPT is text only, so it’s great for searching, analyzing, and planning, but it can’t show you Tallyfy’s visual screens. The setup steps are under For developers.

What you can do

Section titled “What you can do”

For example, you can:

  • Search in plain English: “Find all templates related to employee onboarding that include background check steps.” ChatGPT searches names, descriptions, and step content.
  • Ask it to add the right form fields to collect customer feedback in a process. It suggests the fields and their validation rules.
  • Test an automation idea before you build it. Ask “if I route tasks by deal value, how would these 5 deals get assigned?” and it simulates the result without touching your live processes.
  • Paste an updated procedure and ask ChatGPT to update the matching template. It points out exactly what changed.
  • Have it review the one-off tasks people added to a process last month and suggest which ones belong in the template for good.
  • Ask reporting questions like “Which step in our sales process takes longest, and who’s fastest at it?” The answer links to the exact processes it looked at.

Good to know

Section titled “Good to know”

ChatGPT is a chat window, so anything that needs Tallyfy’s visual screens won’t work well here:

  • No visual tracker. You can’t see Tallyfy’s progress bars, color-coded statuses, or workflow diagrams, so spotting bottlenecks or SLA risk at a glance isn’t possible. For charts and trends, connect a BI tool to Tallyfy Analytics instead.
  • Forms get flattened. Long dropdowns become a wall of text, dates turn into typed text, and upload fields have no file picker.
  • Building templates is clumsy. There’s no drag-and-drop to reorder steps, no visual branching, and no preview.
  • Bulk work is slow. ChatGPT asks you to confirm before it changes many items at once, and big lists come back as pages of text. For bulk changes, use the Tallyfy CLI.
  • It’s not live. ChatGPT only checks when you ask, so you won’t see a teammate’s change until your next query, and two people can’t edit the same template at once.

Which ChatGPT plans you need

Section titled “Which ChatGPT plans you need”

Full read and write access needs a ChatGPT Business, Enterprise, or Edu workspace. Business is the plan OpenAI called Team until August 2025, according to its Business FAQ ↗[1]. On Business, only workspace admins and owners can add Tallyfy. On Enterprise and Edu, an admin can let selected members do it too. Pro accounts can connect with read access only, and nobody can connect from the mobile app. OpenAI keeps these rules in its guide to connecting custom MCP servers ↗[2].

Is it secure?

Section titled “Is it secure?”

You sign in with your own Tallyfy account to authorize the connection and choose which organization to connect, and ChatGPT can only do what that account is allowed to do in Tallyfy. Connect with an account that has only the permissions ChatGPT needs (not a full admin account), review which templates hold sensitive data, and check your access logs now and then. The full setup and security notes are under For developers.

For developers

Section titled “For developers”

(Skip this unless you’re setting up the technical side.)

Server endpoint and support status

Section titled “Server endpoint and support status”

The MCP server is hosted at https://mcp.tallyfy.com/ and exposes the full Tallyfy tool set. ChatGPT connects to it as a custom MCP server (Business, Enterprise, and Edu plans) and through ChatGPT Apps (when you’ve installed it from the apps directory). The server is also published on the Official MCP Registry ↗[3] as com.tallyfy/mcp-server. A verification endpoint at https://mcp.tallyfy.com/.well-known/openai-apps-challenge confirms ownership for OpenAI’s Apps directory; Tallyfy is published in OpenAI’s apps directory as Tallyfy Workflow Automation. MCP itself is governed by the Agentic AI Foundation under the Linux Foundation.

  • Availability - Business, Enterprise, and Edu plans get full read and write access to custom MCP servers, on ChatGPT web only. Pro accounts can connect custom MCP servers with read and fetch permissions only.
  • Access method - A custom MCP server added from Plugins (older screens call it Apps), and ChatGPT Apps from the directory
  • Operations - Full read/write in normal chats on Business, Enterprise, and Edu; read-only in deep research and on Pro
  • Authentication - OAuth 2.1 with PKCE (per RFC 8414) via /.well-known/oauth-authorization-server, with Dynamic Client Registration (RFC 7591)
  • Scopes - Dot notation like mcp.tasks.read, mcp.processes.write, mcp.templates.read
  • Apps SDK - The Tallyfy app from ChatGPT’s apps directory and a custom MCP connection both call this one server, so both read and write the same Tallyfy data. OpenAI’s Apps SDK lets developers build custom interfaces on top of an MCP server.

In the current Plugins screens you don’t need to turn on Developer Mode first. Instead, ChatGPT shows a risk warning for each server you add, and you acknowledge it.

Prerequisites

Section titled “Prerequisites”
  • A ChatGPT Business, Enterprise, or Edu workspace
  • A Tallyfy account with organization access
  • Permission to connect custom MCP servers in that workspace (on Business only admins and owners have it; on Enterprise and Edu, admins can grant it to selected members)

Set up Tallyfy in ChatGPT

Section titled “Set up Tallyfy in ChatGPT”

These steps follow OpenAI’s guide to connecting custom MCP servers ↗[4]. OpenAI updates that guide often, so check it if a label on your screen doesn’t match.

  1. Open Plugins in ChatGPT

    In ChatGPT on the web, go to Plugins. On older screens, admins and owners go to Workspace settings, then Apps, then Create. Other authorized members go to their own settings, then Apps, then Create.

  2. Add Tallyfy as a custom MCP server

    Select the plus button, then Add custom MCP server. Enter a name, such as Tallyfy, and under Connection enter the MCP server URL:

    https://mcp.tallyfy.com

    Choose OAuth for authentication. ChatGPT discovers Tallyfy’s OAuth endpoints automatically through the /.well-known/oauth-authorization-server metadata endpoint (RFC 8414). Review the risk warning, select I understand and want to continue, then select Create as a plugin. On older screens, click Scan Tools instead, sign in when prompted, then click Create.

  3. Authenticate via OAuth 2.1

    ChatGPT initiates OAuth 2.1 with PKCE using Dynamic Client Registration (RFC 7591). You’ll be redirected to Tallyfy’s login page to authorize access.

    During authorization, you’ll select which organization to connect and approve scopes like mcp.tasks.read, mcp.processes.write, etc. If your account isn’t part of a Tallyfy workspace yet, you’ll see a prompt to create one first, then you’re brought back to finish connecting.

    Security tip - Consider using an account with only the permissions ChatGPT needs rather than a full admin account.

  4. Give ChatGPT instructions (optional)

    Paste instructions like these at the start of a conversation so ChatGPT knows how to work with Tallyfy:

    You are connected to a Tallyfy organization's MCP server. Use this connection to:
    - Search for tasks, processes, and templates
    - Retrieve workflow information
    - Analyze template health and suggest improvements
    - Help users manage their work
    

    Always confirm destructive actions before executing.
    Format responses clearly with relevant details.

  5. Test the connection

    Install the plugin, start a new conversation, type @ in the prompt box, and select it. Then try a simple query:

    "Show me all active processes in Tallyfy"

  6. Publish to workspace (optional)

    For team-wide access, publish the app to your ChatGPT workspace. Only admins and owners can publish. On Enterprise and Edu, admins can also choose which groups get access and which actions the app may take.

Example prompts

Section titled “Example prompts”

Finding overdue tasks

Using Tallyfy, find all overdue tasks assigned to the marketing team and summarize them by priority.

ChatGPT looks up the tasks with search_for_tasks, checks their due dates to find the overdue ones, keeps those assigned to the marketing team, and returns a summary grouped by priority.

Analyzing template health

Analyze our "Customer Onboarding" template in Tallyfy and suggest improvements.

Here it pulls the template with get_template, checks it with assess_template_health, reviews its automations with analyze_template_automations, and suggests changes based on what those checks found.

Creating workflow documentation

Generate documentation for our "Invoice Processing" workflow, including all steps and form fields.

It reads the template structure with get_template and writes up every step with its description, form fields, and automations.

These prompts also work well:

Find all templates related to employee onboarding that include background check steps
Add appropriate form fields to collect customer feedback in our support process
If I set up an automation to assign tasks based on deal value, show me how it would route these 5 example deals
Here's our updated SOX compliance procedure. Update our audit template to match these new requirements, highlighting what changed
Analyze one-off tasks added to our hiring processes last month and suggest which should be added to the template
Which step in our sales process has the longest average completion time, and which team members are fastest at completing it?

Limitations of the text-based interface

Section titled “Limitations of the text-based interface”
  • Form field interactions - Dropdowns, multi-selects, and other complex inputs turn into plain text, so a long dropdown arrives as a wall of options to read through. Dates are typed as text, like YYYY-MM-DD HH:MM for a deadline. Upload fields have no file picker: the server’s upload_asset tool can attach a file only from its contents sent as base64 text.
  • Visual workflow representation - ChatGPT can’t draw the Tracker View, with its progress bars and color-coded statuses, or any workflow diagram. You don’t see the step-by-step flow or which steps depend on others.
  • Bulk operations - The MCP server asks you to confirm before it changes many items at once. Use the Tallyfy CLI for bulk work. A large process list comes back as pages of text however you filter or sort it, and batch template updates give no visual confirmation.
  • Real-time collaboration - ChatGPT won’t tell you when a teammate changes something. If they finish a task, you find out the next time you ask. Two people can’t edit the same template at once, either.
  • Template building, saved views, and charts - The drag-and-drop template builder doesn’t come across, so there’s no visual branching and no structure preview. Build templates in Tallyfy itself. Advanced filtering and saved views are missing too, and analytics arrive without charts, graphs, or trend lines; for those, connect your BI tools to Tallyfy Analytics.

ChatGPT is good at natural-language search, what-if automation simulations, and complex questions about your processes.

Security considerations

Section titled “Security considerations”

Use an account with only the permissions the integration needs, or a dedicated service account for ChatGPT. Review which templates and processes contain sensitive information, and audit your data access logs regularly.

Watch for prompt injection, which is text inside a template or task that reads like instructions to the AI. If ChatGPT does something you didn’t expect, check it before you act on it, and report suspicious activity to both OpenAI and Tallyfy support.

Known issues

Section titled “Known issues”
  1. Deep Research is read-only - Deep research can use Tallyfy for read and fetch actions only. Use a normal chat with the Tallyfy plugin selected for write operations.
  2. Context window limits - ChatGPT reads a template as text, and a big one with many steps and fields can be more than it holds at once
  3. No real-time updates - Changes in Tallyfy aren’t reflected until the next query. Live sync between ChatGPT and Tallyfy is planned but not available yet.
  4. Files travel as text - The server’s get_asset_content and upload_asset tools (tool list) read and attach files as base64 text, up to 10MB each. Reading a bigger file returns only its name, type, and size
  5. No webhook support - ChatGPT can’t receive Tallyfy webhooks, so it can’t react on its own when something happens in Tallyfy
  6. Terminology - OpenAI’s help pages now say Plugins, call the Apps screens legacy, and still say connector in places, so other docs may use any of the three names

Best practices

Section titled “Best practices”
  1. Be specific - Instead of “show me tasks,” try “show me high-priority tasks assigned to John due this week”
  2. Batch requests - Put related questions in one prompt, such as “list my overdue tasks and the process each one belongs to”
  3. Finish visual work in Tallyfy - Use ChatGPT to analyze and plan a template change, then make it in Tallyfy’s template builder, where you can drag steps into order and see the branching
Section titled “Related articles”
Get in touch
About Tallyfy