Skip to content

MCP · for developers

Give your AI agent context from real work.

Tydy’s MCP server lets Cursor, Claude and other agents read tasks, chats and files of one workspace and act within the rights you grant.

Endpointhttps://tydy.dev/mcp
AI agent · Tydy MCP
Connected
TelegramChats and files
TydyTasks and context
MCPScoped tools
AI agentCursor, Claude…
→ get_task { task_id: "7f3c…e91" }
← CRM-91 · In progress · 3 messages
→ get_chat_messages { chat_id: "c21a…", limit: 20 }
← 3 messages · screenshot · voice with transcript
→ get_attachment { attachment_id: "a91e…" }
← checkout-mobile.png · short-lived link
→ update_task { task_id: "7f3c…e91", due_at: "2026-10-02" }
← CRM-91 updated

19:05

An agent picks up the task

What it is

An authorised assistant, not a new login

Work stays in Tydy and Telegram. The agent acts as a delegated assistant: it can see task context, move the board and write in chat within your settings.

  1. Reads the task get_task
  2. Reads the chat thread get_task_conversation
  3. Opens the screenshot get_attachment
  4. Moves it to Review move_task
  5. Can't write to Telegram no mcp.telegram.send scope

One workspace per connection

A token or OAuth grant is bound to one workspace and can’t reach any other.

Acts within your rights

The agent never does more than your role allows, even with a broader scope.

Writes are recorded

Changes an agent makes through MCP appear in the workspace history like any other action.

Connect

Two ways to connect

OAuth

For clients that support MCP authorization: sign in to Tydy in the browser and approve access.

  1. 1 Add https://tydy.dev/mcp as a remote MCP server
  2. 2 Sign in to Tydy and pick the workspace
  3. 3 Approve the scopes and the client gets a token

Personal token

Sign in to Tydy and create a personal MCP token (prefix tydy_mcp_) under Settings → MCP, or use OAuth in supported clients. Send Authorization: Bearer <token>.

Cursor · mcp.json
{
  "mcpServers": {
    "tydy": {
      "url": "https://tydy.dev/mcp",
      "headers": { "Authorization": "Bearer tydy_mcp_…" }
    }
  }
}

Leave Client ID and Client Secret empty

Permissions

What an agent can do is an intersection

Access is always bound to one workspace. A token still cannot exceed your role in the team.

Token scopesYour rolePlanEffective access

Effective rights = granted scopes ∩ your live role permissions (RBAC) ∩ plan access. Tokens are stored hashed; revoke takes effect immediately. Attachments use short-lived links.

Scopes

  • mcp.workspace.readView workspace profile
  • mcp.projects.readView projects and columns
  • mcp.tasks.readView tasks
  • mcp.tasks.writeCreate, update, and organize tasks (including subtask trees)
  • mcp.comments.readView comments
  • mcp.comments.writeAdd comments
  • mcp.task_conversation.readRead task conversation
  • mcp.telegram.readView Telegram chats
  • mcp.telegram.sendSend Telegram messages
  • mcp.files.readRead files and attachments
  • mcp.ai_inbox.readView AI Inbox
  • mcp.ai_inbox.reviewReview AI Inbox

Tools

Tools and the scope each one needs

Tool names as the MCP server exposes them.

Tasks & projects13

  • list_projectsmcp.projects.read
  • get_projectmcp.projects.read
  • list_tasksmcp.tasks.read
  • search_tasksmcp.tasks.read
  • get_taskmcp.tasks.read
  • list_task_childrenmcp.tasks.read
  • get_task_treemcp.tasks.read
  • create_taskmcp.tasks.write
  • create_task_treemcp.tasks.write
  • update_taskmcp.tasks.write
  • move_taskmcp.tasks.write
  • add_commentmcp.comments.write
  • get_task_conversationmcp.task_conversation.read

Files1

  • get_attachmentmcp.files.read

Telegram chats6

  • list_telegram_chatsmcp.telegram.read
  • get_chatmcp.telegram.read
  • get_chat_messagesmcp.telegram.read
  • get_chat_messagemcp.telegram.read
  • send_chat_messagemcp.telegram.send
  • reply_to_chat_messagemcp.telegram.send

AI Inbox review5

  • get_ai_inboxmcp.ai_inbox.read
  • get_ai_proposalmcp.ai_inbox.read
  • accept_ai_proposalmcp.ai_inbox.review
  • ignore_ai_proposalmcp.ai_inbox.review
  • restore_ai_proposalmcp.ai_inbox.review

Examples

Prompts you can actually run

  • “Review Victor’s latest messages and update related tasks.”
  • “Open the screenshot he sent and tell me what’s broken.”
  • “Move finished tasks to Review.”
  • “Reply to the client that the fix is deployed.”
Example agent session (abridged)
→ list_tasks { project_id: "5b1e…", priority: "high", limit: 20 }
← 3 tasks · CRM-91 first
→ get_task { task_id: "7f3c…e91" }
← CRM-91 · Fix the pay button on mobile
→ get_chat_messages { chat_id: "c21a…", limit: 20 }
← 3 messages · screenshot · voice with transcript
→ get_attachment { attachment_id: "a91e…" }
← checkout-mobile.png · short-lived link
→ update_task { task_id: "7f3c…e91", due_at: "2026-10-02", description: "…" }
← CRM-91 updated

IDs shortened. Responses are JSON; shown here as one-line summaries.

Connect your agent to Tydy

Create a workspace, then add a token in Settings → MCP.