Connect Your AI Agent to KairoProject

Imagine being able to ask this simple and natural question: "Which project is most at risk right now?" and getting an answer based on your real data, in seconds. That is exactly what connecting KairoProject to an AI agent like Claude or ChatGPT makes possible.

You do not need to be a project management expert to benefit from it. You do not need to be a developer to set it up. All it takes is five minutes and a conversation.

What it really changes

Steer your projects by talking, not searching

Today, making a decision about a project often means navigating multiple screens, cross-referencing data, and exporting a spreadsheet. With an AI agent connected to KairoProject, you ask your question in plain language and get an answer built from your actual data.

No special training required. No dashboard to learn. You talk to your portfolio the way you would talk to a colleague who knows the status of every project inside and out.

Accessible to everyone, not just experts

KairoProject is built on Critical Chain Project Management, a powerful method that takes some learning to master. Connecting an AI agent lowers that barrier: you can ask simple questions ("Is this project on track?") and get a clear interpretation, even if you are not yet fluent in buffers and critical chains.

This is especially valuable for project managers who are new to the method, or for executives who want a quick read without diving into the details.

A full picture, without the effort of consolidating it

Managing several projects in parallel means constantly juggling fragmented information. The AI agent does the synthesis for you: it aggregates the state of your projects, spots warning signals, and presents what matters. You save the time you used to spend pulling things together manually.

Once connected, your AI agent can interact directly with your KairoProject data. No more copy-pasting information or describing the state of your projects by hand.

Real-time visibility

Your assistant accesses the current state of your portfolios, projects, and tasks. It can summarize the situation without you having to open the app.

Decide through conversation

Ask your questions in plain language. The agent analyzes your data, identifies priorities, and helps you make calls without navigating the interface.

Update progress

With the right permissions, it can update the remaining work on a task, which automatically refreshes the CCPM schedule.

Before you start: choosing the right access level

When connecting, you select an access level. This determines what your agent can do. It is not permanent: you can create multiple connections with different levels.

If you are unsure, start with Read-only. You can create a broader-access connection later in under a minute.

The guardrails that protect your data

The CCPM engine remains the single source of truth for every planning calculation: the agent never recomputes anything on its own — it reads values already computed by KairoProject and triggers the same actions available in the interface. Two protections apply systematically, regardless of which assistant you use:

• Pre-action safety check. Before reassigning a resource or modifying a task shared across multiple projects, the agent can check that it isn't degrading a higher-priority project — and warn you explicitly if it is, with the details of which projects would be affected.
• Systematic before/after report. Any action that touches the schedule (progress update, reassignment, recompute) returns the actual observed effect on the plan, not an estimate. The agent can tell you precisely what changed, rather than guessing.

Connect Claude.ai

1. 1

   Go to Settings → Connect an AI assistant. Choose your access level, name the connection (for example: "Claude.ai — read"), then click Create client.

   Copy the Client ID and Client secret that appear immediately: the secret will not be shown again after this step.

2. 2

   Select Claude.ai from the list of assistants, then click Configure callback URL (OAuth). KairoProject automatically registers the required address.

3. 3

   In Claude.ai, go to your account settings and add a new custom connector. Enter the three pieces of information copied from KairoProject: Server URL, Client ID, and Client secret.

4. 4

   Click Connect. A KairoProject window opens: review the requested permissions and click Authorize access. The connection is now active.

Connect ChatGPT

1. 1

   Same process as for Claude.ai: Settings → Connect an AI assistant, choose the access level, create the client, and note your credentials.

2. 2

   In ChatGPT, add a new connector. ChatGPT provides a Callback URL specific to your account.

3. 3

   Return to KairoProject, select ChatGPT, click Configure callback URL (OAuth), and paste the URL provided by ChatGPT. Save.

4. 4

   Finish the configuration in ChatGPT by entering the Server URL, Client ID, and Client secret. The connection is ready to use.

Connect Claude Code

Claude Code is Anthropic's command-line assistant. The connection uses a direct token — no graphical interface required.

1. 1

   Settings → Connect an AI assistant, choose the access level, and create the client.

2. 2

   Select Claude Code from the list. KairoProject displays a ready-to-use command with your token already included.

3. 3

   Paste and run the command in a terminal. Then verify the connection by typing /mcp in Claude Code.

Connect a generic MCP client

For any other MCP (Model Context Protocol) compatible assistant not named above, the connection uses a raw token rather than a guided flow.

1. 1

   Settings → Connect an AI assistant, choose the access level, and create the client. Note the Client ID and Client secret.

2. 2

   Call POST /api/public/oauth/token with grant_type: client_credentials, your client_id, and your client_secret to obtain an access token.

3. 3

   Enter the server URL (https://app.kairoproject.com/api/mcp) and pass the token in the x-kairoproject-token header of your client's JSON configuration.

Real-world usage examples

Once connected, here is what you can ask your agent in plain language.

The agent understands these questions in natural language and acts directly on your data, within its granted permissions. No copy-pasting, no navigating the interface — you just talk to your projects.

Best practices

Create one connection per use case. Rather than one connection with all permissions, create a "Read-only" connection for your weekly reviews and a "Project steering" connection for in-sprint updates. This limits risk in case of an error.

Name your connections clearly. "Claude.ai — portfolio review" or "ChatGPT — task updates": you will easily find which connection does what in your settings.

Revoke what you no longer use. If you tested a connection and no longer need it, revoke it from Settings → Connect an AI assistant. This does not delete any history, but access becomes inactive immediately.

Check permissions before sharing. In an organization, each member can create their own connections. Access tokens are personal: a token created by one member only grants access to data that member can already see.

Known limitations

• No data is cached on the MCP server side: every call re-reads the current state of your data. Your agent may therefore show different information between two messages if the plan changed in between (a recompute triggered elsewhere, by another team member).
• Deferred tool loading can make an existing capability temporarily invisible to the assistant — see the note above about explicitly searching for a tool before concluding it doesn't exist.
• A revoked connection loses access immediately, even for access the agent was actively using.

Frequently asked questions

Read next

• KairoProject Public API — for teams who want to go further and connect their own tools via the REST API.
• What is Critical Chain? — understand the planning methodology behind KairoProject.
• Managing a Project Portfolio — get the most out of the portfolio view, with or without an AI agent.