Temporal Cortex MCP

v0.5.2 · February 2026 · Changelog · Website: temporal-cortex.com

Temporal Cortex is a Model Context Protocol server that gives AI agents deterministic calendar capabilities — temporal context, datetime resolution, multi-calendar availability merging across Google Calendar, Microsoft Outlook, and CalDAV, and conflict-free booking with Two-Phase Commit. Powered by Truth Engine. Install: npx @temporal-cortex/cortex-mcp.

Why do AI agents fail at calendar tasks?

Even the latest LLMs — GPT-5, Claude, Gemini — score below 50% on temporal reasoning tasks (OOLONG benchmark). Earlier models scored as low as 29% on scheduling and 13% on duration calculations (Test of Time, ICLR 2025). Ask "Schedule for next Tuesday at 2pm" and it picks the wrong Tuesday. Ask "Am I free at 3pm?" and it checks the wrong timezone. Then it double-books your calendar.

Most Calendar MCP servers are thin CRUD wrappers that pass these failures through to a single calendar provider — no temporal awareness, no conflict detection, no safety net.

What makes Temporal Cortex different from other calendar MCP servers?

• Temporal awareness — Agents call get_temporal_context to know the actual time and timezone. resolve_datetime turns "next Tuesday at 2pm" into a precise RFC 3339 timestamp. No hallucination.
• Atomic booking — Lock the time slot, verify no conflicts exist, then write. Two agents booking the same 2pm slot? Exactly one succeeds. The other gets a clear error. No double-bookings.
• Computed availability — Merges free/busy data across multiple calendars into a single unified view. The AI sees actual availability, not a raw dump of events to misinterpret.
• Deterministic RRULE expansion — Handles DST transitions, BYSETPOS=-1 (last weekday of month), EXDATE with timezones, leap year recurrences, and INTERVAL>1 with BYDAY. Powered by Truth Engine, not LLM inference.
• Token-efficient output — TOON format compresses calendar data by ~40% fewer tokens than standard JSON, reducing costs and context window usage. TOON is the default output format for all data tools (list_calendars, list_events, find_free_slots, expand_rrule, get_availability). JSON is available via explicit format: "json".

What do I need to run Temporal Cortex?

• Node.js 18+ (for npx to download and run the binary) or Docker
• At least one calendar provider:
  • Google Calendar — requires Google OAuth credentials
  • Microsoft Outlook — requires Azure AD app registration (MICROSOFT_CLIENT_ID)
  • CalDAV (iCloud, Fastmail, etc.) — requires an app-specific password

How do I install Temporal Cortex?

The fastest way to get started:

npx @temporal-cortex/cortex-mcp setup

The cortex-mcp setup wizard walks you through provider authentication, timezone configuration, and MCP client setup interactively. See the First Run Guide for a detailed walkthrough.

Or set up manually in 3 steps:

1. Install prerequisites — Node.js 18+ (or Docker) and at least one calendar provider (Google Calendar, Microsoft Outlook, or CalDAV).
2. Add the MCP configuration to your AI client's config file (see client-specific examples below).
3. Run the auth flow — npx @temporal-cortex/cortex-mcp auth google (or outlook / caldav). This authenticates and configures timezone, week start, and telemetry preferences.

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "temporal-cortex": {
      "command": "npx",
      "args": ["-y", "@temporal-cortex/cortex-mcp@0.5.2"],
      "env": {
        "GOOGLE_CLIENT_ID": "your-client-id.apps.googleusercontent.com",
        "GOOGLE_CLIENT_SECRET": "your-client-secret",
        "TIMEZONE": "America/New_York"
      }
    }
  }
}

Cursor

Add to Cursor's MCP settings (~/.cursor/mcp.json):

{
  "mcpServers": {
    "temporal-cortex": {
      "command": "npx",
      "args": ["-y", "@temporal-cortex/cortex-mcp@0.5.2"],
      "env": {
        "GOOGLE_CLIENT_ID": "your-client-id.apps.googleusercontent.com",
        "GOOGLE_CLIENT_SECRET": "your-client-secret",
        "TIMEZONE": "America/New_York"
      }
    }
  }
}

Windsurf

Add to Windsurf's MCP config (~/.codeium/windsurf/mcp_config.json):

{
  "mcpServers": {
    "temporal-cortex": {
      "command": "npx",
      "args": ["-y", "@temporal-cortex/cortex-mcp@0.5.2"],
      "env": {
        "GOOGLE_CLIENT_ID": "your-client-id.apps.googleusercontent.com",
        "GOOGLE_CLIENT_SECRET": "your-client-secret",
        "TIMEZONE": "America/New_York"
      }
    }
  }
}

Docker

docker run --rm -i \
  -e GOOGLE_CLIENT_ID="your-client-id.apps.googleusercontent.com" \
  -e GOOGLE_CLIENT_SECRET="your-client-secret" \
  -e TIMEZONE="America/New_York" \
  -v ~/.config/temporal-cortex:/root/.config/temporal-cortex \
  cortex-mcp

Build the image first: docker build -t cortex-mcp . (or build directly from the repo: docker build -t cortex-mcp https://github.com/temporal-cortex/mcp.git).

Need help with provider credentials? See the setup guides: Google Calendar, Microsoft Outlook, CalDAV (iCloud/Fastmail). For a complete reference of all environment variables and configuration options, see the Configuration Guide.

How do I authenticate with calendar providers?

The easiest path is npx @temporal-cortex/cortex-mcp setup, which handles authentication, configuration, and MCP client setup in one guided flow. For individual provider auth, run the commands below:

# Google Calendar (default)
npx @temporal-cortex/cortex-mcp auth google

# Microsoft Outlook
npx @temporal-cortex/cortex-mcp auth outlook

# CalDAV (iCloud, Fastmail, or custom server)
npx @temporal-cortex/cortex-mcp auth caldav

# Docker (interactive auth — needs terminal + browser)
docker run --rm -it \
  -e GOOGLE_CLIENT_ID="your-id" -e GOOGLE_CLIENT_SECRET="your-secret" \
  -p 8085:8085 \
  -v ~/.config/temporal-cortex:/root/.config/temporal-cortex \
  cortex-mcp auth google

Each auth flow saves credentials to ~/.config/temporal-cortex/credentials.json and registers the provider in ~/.config/temporal-cortex/config.json. You can connect multiple providers — the server discovers all configured providers on startup and merges their calendars into a unified view.

During auth, the server guides you through interactive setup:

• Timezone — auto-detects your system timezone and opens a fuzzy-search picker with all 597 IANA timezones (type to filter, arrow keys to navigate)
• Week start — arrow-key selection between Monday (ISO standard) and Sunday
• Telemetry — optional anonymous usage data (default: off)

All preferences are stored in ~/.config/temporal-cortex/config.json and used by all temporal tools. You can override them per-session with the TIMEZONE and WEEK_START env vars.

After authentication, verify it works by asking your AI assistant: "What time is it?" — the agent should call get_temporal_context and return your current local time.

For a guided workflow, install the calendar-scheduling Agent Skill to teach your AI agent the orient-resolve-query-book pattern.

Temporal Cortex Platform

Instead of running the MCP server locally via npx, you can use the managed Temporal Cortex Platform. No Node.js installation or local OAuth credentials required.

Getting started:

1. Sign up at app.temporal-cortex.com.
2. Connect your Google Calendar or Microsoft Outlook account via OAuth in the dashboard.
3. Generate an API key from the dashboard.
4. Add the MCP config to your AI client (see examples below).

Claude Desktop (Platform)

Add to your Claude Desktop config file:

{
  "mcpServers": {
    "temporal-cortex": {
      "url": "https://mcp.temporal-cortex.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Cursor (Platform)

Add to Cursor's MCP settings (~/.cursor/mcp.json) using the same format:

{
  "mcpServers": {
    "temporal-cortex": {
      "url": "https://mcp.temporal-cortex.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Platform capabilities (beyond Local Mode):

• No OAuth credentials to manage -- calendar connections are handled in the dashboard via standard OAuth flows.
• No Node.js required -- the client connects directly to the cloud endpoint over HTTP.
• Usage dashboard -- monitor tool calls, connected calendars, and billing from app.temporal-cortex.com.
• Managed calendar connections -- token refresh, re-authentication, and provider health are handled server-side.
• Multi-agent coordination -- distributed locking prevents double-bookings when multiple agents schedule simultaneously.
• Usage metering -- track tool calls per agent and team from the dashboard.
• Content firewall -- automatic prompt injection detection and zero-width Unicode stripping.
• Caller-based policies -- enforce booking rules per agent (max duration, allowed hours, booking limits).

All 12 tools and 4 layers work identically. The Platform adds safety, coordination, and visibility infrastructure on top.

What tools does Temporal Cortex provide?

Temporal Cortex exposes 12 Model Context Protocol tools organized in 4 layers:

Layer 1 — Temporal Context

Layer 2 — Calendar Operations

Layer 3 — Availability

Layer 4 — Booking

See docs/tools.md for full input/output schemas and usage examples.

How does Temporal Cortex handle recurrence rules?

Most AI models and calendar tools silently fail on recurrence rule edge cases. Run the challenge to see the difference:

npx @temporal-cortex/cortex-mcp rrule-challenge

5 cases where LLMs consistently fail

1. "Third Tuesday of every month" across DST (March 2026, America/New_York)

The third Tuesday is March 17. Spring-forward on March 8 shifts UTC offsets from -05:00 to -04:00. LLMs often produce the wrong UTC time or skip the month entirely.

2. "Last Friday of every month" (BYSETPOS=-1)

RRULE:FREQ=MONTHLY;BYDAY=FR;BYSETPOS=-1 — LLMs frequently return the first Friday instead of the last, or fail to handle months with 4 vs 5 Fridays.

3. "Every weekday except holidays" (EXDATE with timezone)

EXDATE values with explicit timezone offsets require exact matching against generated instances. LLMs often ignore EXDATE entirely or apply it to the wrong date.

4. "Biweekly on Monday, Wednesday, Friday" (INTERVAL=2 + BYDAY)

RRULE:FREQ=WEEKLY;INTERVAL=2;BYDAY=MO,WE,FR — The INTERVAL=2 applies to weeks, not individual days. LLMs frequently generate every-week occurrences instead of every-other-week.

5. "February 29 yearly" (leap year recurrence)

RRULE:FREQ=YEARLY;BYMONTH=2;BYMONTHDAY=29 — Should only produce instances in leap years (2028, 2032...). LLMs often generate Feb 28 or Mar 1 in non-leap years.

Truth Engine handles all of these deterministically using the RFC 5545 specification. No inference, no hallucination.

How does the MCP server architecture work?

The MCP server is a single Rust binary distributed via npm and Docker. It runs locally on your machine and communicates with MCP clients over stdio (standard input/output) or streamable HTTP.

Request flow in 4 stages:

1. Receive — The server accepts JSON-RPC messages from MCP clients over stdio (default) or streamable HTTP (when HTTP_PORT is set).
2. Resolve — Truth Engine converts human datetime expressions into precise UTC timestamps, handling timezone conversion and DST transitions deterministically.
3. Compute — For availability queries, the engine merges free/busy data across all connected calendar providers. For RRULE expansion, it generates concrete instances following RFC 5545 rules.
4. Execute — For booking operations, Two-Phase Commit acquires a lock, verifies the slot is free, writes the event, and releases the lock. Failure at any step triggers rollback.

TOON (Token-Oriented Object Notation) compresses calendar data for LLM consumption — ~40% fewer tokens than JSON, with perfect roundtrip fidelity. TOON is the default output format for all data tools; use format: "json" when you need structured JSON.

Stdio vs HTTP Transport

Transport mode is auto-detected — set HTTP_PORT to switch from stdio to HTTP.

• Stdio (default): Standard MCP transport for local clients (Claude Desktop, VS Code, Cursor). The server reads/writes JSON-RPC messages over stdin/stdout.
• HTTP (when HTTP_PORT is set): Streamable HTTP transport per MCP 2025-11-25 spec. The server listens on http://{HTTP_HOST}:{HTTP_PORT}/mcp with SSE streaming, session management (Mcp-Session-Id header), and Origin validation. Requests with an invalid Origin header are rejected with HTTP 403.

# HTTP mode example
HTTP_PORT=8009 npx @temporal-cortex/cortex-mcp@0.5.2

Local Mode vs Platform Mode

Mode is auto-detected — there is no configuration flag.

• Local Mode (default): No infrastructure required. Uses in-memory locking and local file credential storage. Supports multiple calendar providers (Google, Outlook, CalDAV) with multi-calendar availability merging. Designed for individual developers.
• Platform Mode (managed service at mcp.temporal-cortex.com): Managed multi-tenant hosting with Postgres-backed credentials, Bearer token authentication, guardrails (rate limiting + booking caps), usage metering, and distributed locking for multi-agent safety. Designed for teams and production deployments.

How do I configure Temporal Cortex?

At least one calendar provider must be configured. See the provider setup guides: Google Calendar, Microsoft Outlook, CalDAV (iCloud/Fastmail). For a complete configuration reference, see the Configuration Guide.

How do I troubleshoot common issues?

See provider-specific troubleshooting: Google Calendar, Microsoft Outlook, CalDAV.

Frequently Asked Questions

Does Temporal Cortex work without internet access?

Layer 1 tools (temporal context, datetime resolution, timezone conversion, duration computation, timestamp adjustment) are pure computation and need no network access. Calendar tools (Layers 2-4) require network access to reach Google Calendar, Microsoft Outlook, or CalDAV APIs. The MCP server itself runs locally on your machine.

Which AI clients are supported?

Any Model Context Protocol-compatible client works. Tested configurations are provided for Claude Desktop, Claude Code, VS Code with GitHub Copilot, Cursor, and Windsurf. The server uses stdio transport by default and also supports streamable HTTP transport for custom integrations.

Can I connect multiple calendar providers simultaneously?

Yes. Run the auth flow for each provider (Google, Outlook, CalDAV) separately. The server discovers all configured providers on startup and merges their calendars into a unified availability view. Use provider-prefixed IDs like google/primary or outlook/work to target specific calendars.

How does Two-Phase Commit prevent double-bookings?

When book_slot is called, the server acquires a time-range lock, verifies no conflicting events or active locks exist, writes the event to the calendar API, then releases the lock. If any step fails, the operation rolls back. Two agents booking the same slot simultaneously will have exactly one succeed and the other receive a clear error.

What is TOON and why does it reduce costs?

TOON (Token-Oriented Object Notation) compresses calendar payloads by ~40% fewer tokens compared to JSON (38% on a Google Calendar event schema benchmark) while maintaining perfect roundtrip fidelity. Fewer tokens means lower API costs and more room in the LLM context window. TOON is now the default output format for all data tools (list_calendars, list_events, find_free_slots, expand_rrule, get_availability). Use format: "json" to get JSON output instead.

How does Temporal Cortex handle daylight saving time?

All temporal tools are DST-aware. adjust_timestamp with "+1d" across a spring-forward boundary preserves wall-clock time (1:00 AM EST becomes 1:00 AM EDT). RRULE expansion keeps recurring events at their local time across DST transitions. get_temporal_context reports whether DST is currently active and predicts the next DST transition — including the transition date, direction (spring-forward or fall-back), and days until it occurs.

What is the difference between Local Mode and Platform Mode?

Local Mode (default) runs on your machine with in-memory locking, local file credential storage, and no infrastructure required — all 12 tools work with zero setup. Platform Mode (at mcp.temporal-cortex.com) adds managed OAuth lifecycle, multi-agent coordination with distributed locking, usage metering, caller-based policies, a content firewall, and a dashboard UI. Both expose the same 12 tools and 4 layers — the Platform adds safety, coordination, and visibility for teams.

How bad are LLMs at temporal reasoning?

Even the latest frontier models — GPT-5, Claude Sonnet 4, Gemini 2.5 Pro — score below 50% on temporal reasoning tasks (OOLONG benchmark, arXiv:2511.02817). Earlier models scored as low as 29% on scheduling and 13% on duration calculations (Test of Time, ICLR 2025, arXiv:2406.09170). Temporal questions are consistently the most challenging category for LLMs. Temporal Cortex replaces LLM inference with deterministic computation for all calendar math.

Is there a managed cloud option?

Yes. The Temporal Cortex Platform is available at app.temporal-cortex.com. Sign up for free to get managed hosting, OAuth lifecycle management, and a usage dashboard. Pro and Enterprise tiers add Open Scheduling, caller-based policies, and compliance features.

Does Temporal Cortex collect telemetry?

During setup, an interactive prompt asks if you'd like to share anonymous usage data (default: off).

Collected: tool names, success/error counts, platform, version. Never collected: calendar data, events, or personal info.

Non-interactive sessions (MCP stdio) auto-opt-out. Change your choice anytime:

export TEMPORAL_CORTEX_TELEMETRY=off

How do I teach my AI agent the scheduling workflow?

The calendar-scheduling Agent Skill teaches AI agents the correct workflow for using these tools — from temporal orientation through conflict-free booking. Install it to give your agent procedural knowledge for calendar operations:

# Claude Code
npx skills add temporal-cortex/skills

The skill follows the Agent Skills specification and works with Claude Code, OpenAI Codex, Google Gemini, GitHub Copilot, Cursor, and 20+ other platforms.

What is the computation layer behind Temporal Cortex?

The computation layer is open source:

• temporal-cortex-core — Truth Engine (temporal resolution, RRULE expansion, availability merging, timezone conversion) + TOON (token compression)
• Available on crates.io, npm, and PyPI
• 510+ Rust tests, 42 JS tests, 30 Python tests, ~9,000 property-based tests

Contributing

Bug reports and feature requests are welcome. See CONTRIBUTING.md.

License

MIT