Connect AI apps with MCP

The MONOid MCP connector lets an AI app search and manage your MONOid data through natural language. You do not need to install the MONOid CLI.

Connect

1. Open your AI app’s connector settings and add a remote MCP server.
2. Enter https://api.usemonoid.com/mcp.
3. Choose OAuth authentication, then sign in to MONOid.
4. Review the app name and choose its access level.
5. Allow the connection, then enable the connector in your conversation if your AI app requires it.

In Claude, start in Customize → Connectors → Add custom connector. Interface labels may differ between client versions.

The connector uses standard MCP over Streamable HTTP with OAuth and PKCE S256. Use your client’s published identity (Client ID Metadata Document, or CIMD) when available; dynamic client registration is also supported. The current flow uses public clients without a client secret. Clients that require client-secret authentication, private-key authentication, or local MCP processes need an additional supported integration path. PDF attachments and approval interfaces depend on the AI app you use.

Use OAuth for this endpoint. Existing MONOid CLI API keys cannot authenticate an MCP connection.

Choose access

The consent screen only allows permissions the client requested. If a level is unavailable, reconnect from the client with the required permissions. Increasing permissions requires new consent.

Each connection uses your own MONOid account. It covers your personal workspace and every organisation you can access, including organisations you join later. Existing membership and record permissions still apply: the connector cannot grant access you do not already have.

Ask about your day

What are my tasks today? Include tasks scheduled today and tasks due today,
without duplicates. Show overdue tasks separately, with their projects and
routine blocks.

Your MONOid timezone determines today. Set it in your account preferences, or provide a timezone when the app asks. The connector should not infer your timezone from the server’s location.

The today view includes active work; completed and cancelled tasks are omitted. Overdue means a deadline before today. Large result sets have separate pagination for scheduled, due, and overdue tasks, so the AI should follow each stream before claiming a complete list.

You can also ask the app to find a project, inspect a routine, update task progress, or retrieve your day or week plan. When several organisations contain matching names, identify the intended destination before making changes.

Turn a document into a plan

Attach the PDF in your AI app, then ask:

Read this PDF and break the work into containers, projects, and tasks in
MONOid. Check my existing structure and routine blocks. Propose the plan
and schedule first, including any estimated durations or missing deadlines.

The AI app reads the document; MONOid provides the records and scheduling tools. Once you approve the proposed plan—or explicitly ask the app to create it immediately—the app can create the structure and schedule tasks into dated routine blocks.

Tell the app whether existing commitments may move when it needs more time. Ask it to surface missing information and capacity conflicts. A routine template describes a reusable pattern; scheduled tasks belong to a dated routine block. The connector supports resolving a template to the appropriate date.

For a large operation, ask for a progress summary containing created record IDs and any failed steps. If an operation fails or the connection drops, inspect the existing records before retrying. Separate tool calls are not one atomic transaction.

Container, project, and task creation and task upsert accept a stable requestId for each intended step. Repeating the same completed request within 24 hours returns its saved result. If a response says the outcome is uncertain, inspect the records instead of trying a fresh request ID. This protection does not make an entire plan atomic or cover every write operation.

CLI and MCP

MCP exposes MONOid product operations through named tools with typed arguments. It covers the same product areas as the CLI: task and project management, organisations and teams, routines, reviews and notes, labels, calendars, integrations, settings, and agent runs. Planning and search tools help the AI obtain context before changing records.

CLI installation, terminal output formats, and local credential profiles are not remote product operations. MCP uses the connection’s account and consent instead. Tool availability also depends on the connection’s permissions.

Deletion is available with full access. Before deleting a container or project, identify the exact record and understand which related records the operation affects. Permission to delete does not resolve an ambiguous request.

Revoke a connection

Open MONOid Settings → Connected apps to see each client’s access level, connection date, and last use. Select Revoke access to stop subsequent requests and token refreshes. Changes already completed in MONOid remain.

Removing the connector in an AI app and revoking its MONOid grant are separate actions. Revoke in MONOid when you want to ensure that the grant can no longer be used.

Troubleshooting

• Sign-in fails: restart the connection from your AI app. Authorization codes expire; callbacks must match the client’s registered URL.
• A tool is unavailable or returns a permission error: check the connection’s access level and your current organisation permissions. Reconnect if you need a higher access level.
• An organisation is missing: confirm you connected the intended MONOid account and that it still has access.
• The app cannot read a PDF: use a client with attachment support or provide the extracted text. The connector does not parse local files.
• The connector cannot be reached: remote clients need HTTPS access to the hosted endpoint. A local CLI command is not an MCP server URL.