Connect Assistants

This guide shows you how to connect ChatGPT, Claude, Cursor, or another compatible assistant to the right Agiflow resource scope.

Quick answer

Choose the connection path for your client: ChatGPT app/plugin, Claude remote connector, or coding-agent MCP. Use generated endpoints, authorize the smallest useful scope, and verify it with a read-only request.

Maintained by Agiflow. Published 19 May 2026. Last updated 14 Oct 2026.

01 : Choose Your Client

Start in Agiflow's Connect AI tools panel. Choose your client and follow its setup instructions. An available ChatGPT app/plugin, a Claude remote connector, and a coding-agent MCP connection are different setup paths.

ChatGPT app/plugin

Open Settings, Plugins, select Agiflow, then install and connect when prompted. Workspace policy may restrict availability.

Claude remote connector

Open Customize, Connectors and add a custom web connector using Agiflow's generated MCP App URL. Desktop and web use the remote connection path.

Coding agent or custom client

Use the matching guide for Claude Code, Cursor, Codex, or VS Code. Their configuration files are not interchangeable.

Follow the ChatGPT guide, Claude guide, or coding-agent guide. Custom MCP connections in ChatGPT are an advanced path with separate developer-mode and workspace-permission requirements.

💡 Endpoint families, not interchangeable URLs: Coding agents use /api/v1/mcp. App-capable Claude connections use /api/v1/mcp-app/<version>, while the ChatGPT App family uses /api/v1/chatgpt-app/<version>. These are path patterns, not complete configuration snippets. Copy the full URL and current version from Agiflow instead of constructing one.

02 : Choose Scope And Endpoint

Choose the smallest scope that covers the intended work. Use a project for one board, or a work unit or task for a narrower handoff. Use organization scope only when the assistant actually needs workspace-wide work. Review the scope offered by the connection panel and authorization screen.

💡 Recommended: generated OAuth connection: Copy the generated client endpoint and complete browser authorization. A manually copied Organization ID and an API-key header are not required for this path. For a narrower direct connection, use the relevant project-board connection panel rather than guessing resource IDs.

Advanced reference: the following are direct scoped API path patterns. They are not a universal JSON configuration and do not grant access by themselves. Use the generated full URL and the authentication method supported by that connection.

Organization

/api/v1/organizations/YOUR_ORG_ID/mcp

Use when the assistant should help across the workspace.

Project

/api/v1/organizations/YOUR_ORG_ID/projects/YOUR_PROJECT_ID/mcp

Use for one project board and its tasks, work units, artifacts, vault entries, and workflows.

Work unit

/api/v1/organizations/YOUR_ORG_ID/projects/YOUR_PROJECT_ID/work-units/YOUR_WORK_UNIT_ID/mcp

Use for one feature, milestone, or epic inside a project.

Task

/api/v1/organizations/YOUR_ORG_ID/tasks/YOUR_TASK_ID/mcp

Use for one task and its comments.

03 : Authorize Access

Browser authorization

Recommended

For OAuth, use the client's connect or authenticate action, sign in, and inspect the consent screen. Approve only the intended organization and resource. The available tools depend on that scope and the interface version; a narrower connection need not expose an organization-wide project list.

API key

For supported headless or API-key connections, create a key in Agiflow and use the generated scoped endpoint. API keys use the x-api-key header, not an OAuth Bearer token. Store the key in the client's secret configuration, never in a public repository, screenshot, or support message.

API-key header fragment, only for clients supporting custom headers

"headers": {
  "x-api-key": "${API_KEY}"
}

04 : Verify The Connection

Ask: "Use the connected Agiflow tools to identify my authorized scope. Do not create, edit, or delete anything."

For organization scope, request the intended project by name and compare its ID with your board.

For project scope, request that project and its statuses. For work-unit or task scope, verify that resource instead of listing every project.

Compare the returned name and ID with Agiflow. Inspect the tools exposed by your connection rather than assuming every version has the same discovery command.

💡 When verification fails: Check that the same account can open the intended resource, that the endpoint matches the chosen client, and that authorization completed. Reconnect only to the intended scope. Do not grant broader access or create duplicate resources to make a check pass. Use the troubleshooting guide and remove secrets from error reports.

Client interfaces change. Consult the official ChatGPT connection guidance, ChatGPT custom MCP requirements, and Claude connector guidance for current availability and interface labels.

Need the exact tool list?

Use the capability reference after setup to see which tools, prompt skills, artifacts, vault entries, and workflow locks are available for each scope.

View capability reference