Plan the workday across calendar, tickets, email, and docs
Posted on Mar 11 • Edited on Mar 13 • Originally published at keycard.ai. Task: The author built "Plan My Today," a dashboard with an embedded agent "Todaygent" that aggregates tasks from Google Calendar, Linear, Gmail, GitHub, Slack, and Granola via MCP servers. The system fetches data concurrently, scores priorities using tuned algorithms, and provides a unified view for daily planning. The human retains full control over task status changes in source apps, using the agent only for reasoning, prioritization, and meeting prep, while managing dismissals and priority adjustments through a local JSON store.; tools: custom-built or unnamed agent. Human role: The original post does not say how a person stayed involved.; limitations: The agent has strictly read-only access to external tools; it cannot modify tasks, send emails, or change calendar events. Authentication relies on Keycard for ephemeral, task-scoped tokens, requiring a specific backend architecture. The system is designed for single-user local use, not multi-tenant production, and lacks deterministic guardrails for complex agent reasoning loops..
Tools
custom-built or unnamed agent
Published on Agent Work Atlas:
Overview
Posted on Mar 11 • Edited on Mar 13 • Originally published at keycard.ai
Workflow
The author built "Plan My Today," a dashboard with an embedded agent "Todaygent" that aggregates tasks from Google Calendar, Linear, Gmail, GitHub, Slack, and Granola via MCP servers. The system fetches data concurrently, scores priorities using tuned algorithms, and provides a unified view for daily planning. The human retains full control over task status changes in source apps, using the agent only for reasoning, prioritization, and meeting prep, while managing dismissals and priority adjustments through a local JSON store.
Reported result
The author reports using the tool daily for personal planning but does not provide measured business outcomes, efficiency metrics, or success rates. The post notes that the agent's reasoning quality requires constant manual tuning of scoring weights and that the author sometimes questions if time is actually saved. No external validation or quantitative results are presented.
Human role
The original post does not say how a person stayed involved.
Limitations
The agent has strictly read-only access to external tools; it cannot modify tasks, send emails, or change calendar events. Authentication relies on Keycard for ephemeral, task-scoped tokens, requiring a specific backend architecture. The system is designed for single-user local use, not multi-tenant production, and lacks deterministic guardrails for complex agent reasoning loops.
Try it
Build a read-only daily planning agent named "Todaygent" that aggregates tasks from [replaceable inputs: e.g., Google Calendar, Linear, Gmail, Slack] using MCP servers. 1. Implement a backend that exchanges a single user JWT for ephemeral, scoped tokens for each provider. 2. Create a tool `list-all-tasks-today` that fetches data concurrently and scores priorities based on [replaceable inputs: custom scoring rules]. 3. Develop a chat interface where the agent reasons about blockers and meeting prep but cannot write to external APIs. 4. Use a local JSON store for user overrides (dismissals, priority changes). 5. Before finalizing, run a dry test with sample data to verify token scoping and read-only constraints. Confirm with the user that no external write actions are attempted. Output a unified dashboard view with priority and source grouping.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.