The Float MCP is currently in beta and considered an experimental release.
As a new and evolving standard, its design and best practices are still being refined. Some connections and contexts remain untested, and responses may not always be fully accurate or reliable. This beta aims to identify the most valuable use cases and gather feedback to improve future versions.
We’d love to hear your thoughts! Share feedback HERE!
Float MCP: Use AI to ask about your team
MCP (Model Context Protocol) is a standardized framework that allows external AI agents to interact with your Float account without any custom development.
The Float MCP server acts as a bridge between Float’s API and AI tools like ChatGPT, Claude (or any other MCP-enabled system), and provides real-time, read-only access to your team’s data.
You can think of MCP as a middleman: it receives your requests from an AI tool, fetches the relevant data from Float, and returns the results directly to your AI interface in natural language.
With MCP, you can ask questions about your team’s availability, projects, tasks, and time off. It’s a powerful tool for Resource Managers to generate summaries, uncover insights faster, and get a clear, up-to-date snapshot of their team’s data.
Example prompts to ask Float
Float MCP is currently read-only and best suited for bite-sized, specific queries. Here are some sample questions you can ask once you're connected.
From simple queries:
Give me the list of my projects.
Who’s on time off next week?
Give me an update on the “Float Website Redesign” project.
What projects is John working on today?
What’s the utilization rate for Jamie this week?
To more complex ones:
Show all team members who are over capacity or on leave in the next month, grouped by role. For each role, list who’s available to cover their work.
Based on the scheduled hours for next month, which active projects are on track to exceed their budgeted hours? Include estimated overage percentages.
What’s the average profit margin across all active projects in Float? List the top three with the highest margins, and the three with the lowest.
Looking at the past quarter, which teams or departments were consistently over capacity (above 80%)? Summarize by week or month.
Best practices
Break down large requests into smaller, specific queries. For example, instead of using a general prompt like:
Find overbooked people and tell me who can cover for them,
Try being more precise and rewrite your request to be more specific, for example:
- List all people in the Marketing department who are scheduled over 100% of their weekly capacity next week.
- Recommend which project could be removed from their Schedule.
- Suggest who could take on that project. A good candidate should belong to the same department, have a similar role, and have enough available capacity to take on the work.Be explicit about the data or format you want.
Cross-check output data for potential inaccuracies.
Be aware that large or complex queries can result in slower performance or AI hallucinations.
MCP setup requirements
Access rights
Only Account Owners and Admins can connect to the Float MCP. If a Manager or Member attempts to access it, they will be notified about insufficient permissions.
Supported tools
The Float MCP Server currently supports ChatGPT, Claude, and Visual Studio Code.
ChatGPT and Claude both require paid accounts to use custom connectors.
Visual Studio Code (VSCode) supports custom connectors on its free tier.
For ChatGPT, Developer Mode must be enabled to connect successfully.
Connecting the Float MCP to your AI
To connect the Float MCP server, you’ll need to add Float MCP as a custom connector in your AI tool. Personal ChatGPT and Claude accounts support custom MCP connectors by default. For Team or Business AI workspaces, additional permissions may be required. Workspace admins can choose to restrict or allow access to MCP connectors based on internal security and data-access policies. To connect the Float MCP to a team workspace, a workspace admin must enable Developer Mode before anyone in the workspace can add custom connectors.
ChatGPT
To connect the Float MCP to ChatGPT:
1) Navigate to Settings > Apps & Connectors > Advanced settings and turn on the Developer Mode
2) Navigate back to Apps & Connectors and click Create to add a custom connector.
3) Name the connector (for example, Float MCP), enter https://mcp.float.com/mcp in the MCP Server URL field, review and agree to the security notice, and hit Create. Optionally, you can also add an icon and a description of the connector.
📝Note: If you are on a Team or Business GPT plan, one of your team’s admins needs to enable Developer Mode and add the custom Float MCP connector first. After that, you will see it on the connectors list and will be able to connect right away.
4) If you are a member of multiple Float teams, you will be prompted to select the one you want to access. Select the team from the drop-down and hit Allow to grant the Float MCP Server permission to access your Float information.
By clicking Allow during authentication, you grant this ChatGPT access to retrieve read-only data from your Float workspace. It can only access data that you explicitly request within your conversations. It cannot directly access your Float account or modify your workspace independently. This access is secured via OAuth 2.0 and limited to Account Owners and Admins only. Your authorization will automatically expire if your access rights change.
📝Note: To improve MCP performance, Float captures a short, non-personal summary of your prompt intent; no specific or sensitive information is collected. Float does not control how ChatGPT stores retrieved data. Please review their privacy policies before connecting.
5) Once connected, Float MCP will show under enabled connectors.
6) To start using MCP, open a new chat and select the Float MCP from the tools list.
7) You are all set! With the connector enabled, you can now ask ChatGPT questions about your Float team.
8) ChatGPT will use the MCP to fetch the requested data for you and engage in the conversation.
📝Note: Even though Float MCP is read-only, ChatGPT currently classifies all MCP connector tool calls as “write” actions to ensure security and user consent. That’s why you may see a warning asking you to confirm that you acknowledge it. You can choose to remember it for the whole conversation.
Claude
To connect the Float MCP to Claude:
1) Navigate to Settings > Connectors > Add custom connector.
2) Name the connector (for example, Float MCP), enter https://mcp.float.com/mcp in the Remove MCP server URL field, and hit Add.
3) Float MCP will now show on the connectors list. To activate it, click Connect.
4) If you are a member of multiple Float teams, you will be prompted to select the one you want to access. Select the team from the drop-down and hit Allow to grant the Float MCP Server permission to access your Float information.
By clicking Allow during authentication, you grant this Claude access to retrieve read-only data from your Float workspace. It can only access data that you explicitly request within your conversations. It cannot directly access your Float account or modify your workspace independently. This access is secured via OAuth 2.0 and limited to Account Owners and Admins only. Your authorization will automatically expire if your access rights change.
📝Note: To improve MCP performance, Float captures a short, non-personal summary of your prompt intent; no specific or sensitive information is collected. Float does not control how Claude stores retrieved data. Please review their privacy policies before connecting.
5) Once connected, Float MCP will show under enabled connectors.
6) To start using MCP, open a new chat and ensure Float MCP is enabled on the tools list.
7) You are all set! With the connector enabled, you can now ask Claude questions about your Float team.
8) Claude will use the MCP to fetch the requested data for you and engage in the conversation.
Disconnecting Float MCP
ChatGPT
To disconnect Float MCP in ChatGPT, navigate to Settings > Apps & Connectors > Float MCP > Disconnect. To completely remove the connector from your GPT account, select Delete.
Claude
To disconnect the Float MCP in Claude, navigate to Settings > Connectors and select Disconnect. To completely remove the connector from your Claude account, select Delete.
Security
At Float, protecting your data is our top priority. The Float MCP is built with robust security measures to ensure your workspace data remains safe and accessible only to authorized users.
All communication is encrypted via HTTPS (TLS 1.2 or later).
OAuth 2.0 ensures secure identity verification and access control.
Only Account Owners and Admins can connect their workspace to the Float MCP.
Automatic access revocation: If an Account Owner’s and Admins’ access rights are downgraded, their MCP access remains active only until their token expires (which takes around one hour).
MCP uses Float’s public API in a read-only mode. It can retrieve data (e.g., projects, budgets, rates) but cannot modify it.
Data privacy
Float does not control how connected third-party tools, such as ChatGPT or Claude, store or use retrieved data. To help protect your information:
Use business or approved LLM accounts, not personal ones.
Turn off data sharing for model training in your LLM settings.
Avoid sending Float data to public or shared services (such as Notion or Google Sheets) without verifying access permissions.
Best practices for secure use:
Connect Float MCP only from verified admin accounts.
Keep production data separate from experimental or personal LLM environments.
Audit admin access regularly and remove temporary permissions after testing.
MCP limitations
Float MCP is currently in beta and may show limited performance with:
Complex, multi-step tasks
Large data volumes (e.g., reporting across all projects)
Company-wide insights for larger organizations
Provided data is read-only. Write/post/modify operations are not supported.
Switching teams within a session isn’t supported. To access another team, disconnect the MCP and sign in again, selecting the desired team during the authentication process.
API rate limit is 200 per minute.
Enabling ChatGPT's developer mode disables chat history. This is a limitation of ChatGPT, not Float MCP's.
Performance depends on the AI model used (e.g., GPT-4o-mini is less reliable than GPT-5 or Claude Sonnet 4.5).
Accuracy depends on the AI's interpretation and the volume of data. You can improve accuracy by following our best practices: be explicit in your requests and, when asking for calculations, provide the exact formula directly to your AI. Large datasets reduce both performance and accuracy. This cannot be fully solved for. To improve results, avoid fetching all data at once and narrow your queries whenever possible.
Troubleshooting common issues
“Your site admin must authorize this app” error - Ensure your organization’s Admin has approved Float MCP access.
App not appearing in Connected Apps - Verify you’re signed in to the correct Float account.
Permission issues - Your authorization token reflects your access rights in Float.
If access is limited, confirm your permissions with the Account Owner or an Admin.If you are experiencing other issues, try deleting the connector and then recreating it. If the issue persists, contact our support team at support@float.com.