Scheduled Chats

Scheduled Chats are experimental and off until you opt in. Scheduled runs use model, tool, and Code Interpreter resources without anyone watching, so check the deployment requirements below before enabling them.

Enable

Add interface.schedules to librechat.yaml and restart LibreChat:

interface:
  schedules:
    use: true
    create: true
    maxPerUser: 10
    minIntervalMinutes: 60
    autoDisableAfterFailures: 5
    fireConcurrency: 5
    requireProject: false
    # projectId: '000000000000000000000000'

use and create seed the SCHEDULES permissions for the built-in USER role at startup; after that, manage them per role in the Admin Panel. use lets a user see their schedules. create also lets them create, edit, enable, run, and delete them. The other keys are runtime limits:

schedules: true enables the feature with these defaults. schedules: false, or use: false, is a deployment-wide stop that role, group, and user overrides cannot lift.

Project rules are rechecked at every run. A schedule that no longer satisfies requireProject, or whose project was deleted, disables itself instead of starting an unscoped conversation. Chat Projects belong to one user, so a deployment-wide projectId only works for that project's owner; set it in role or user overrides when owners need different destinations.

The full field list, including admissionConcurrency, mcpPreflightConcurrency, and mcpPreflightTimeoutMs for bounding admission and MCP readiness checks separately from generation, is in interface.schedules.

Unattended MCP access

Before a run starts, LibreChat checks every MCP tool the scheduled agent and its subagents need. A run that would need an interactive login does not start; the schedule shows which agent needs attention.

To let users authorize MCP tools for unattended runs, turn on interface.schedules.mcpConsent:

• mcpConsent.enabled turns the feature on.
• mcpConsent.maxLifetimeHours caps how long an authorization lasts, from 1 to 8760 hours.
• mcpConsent.resources lists the MCP resources that may be authorized.
• mcpConsent.readOnlyPolicy lists trusted read-only tool declarations, keyed by server and upstream tool name.
• mcpReceiptRetry.baseMs and mcpReceiptRetry.maxMs set the backoff for retrying consent receipts.

Users then see a Scheduled MCP authorization section in the schedule editor. An authorization covers only the listed agent and resources until its deadline and never permits writes.

Deployment requirements

Scheduled Chats coordinate through MongoDB leases, so several replicas can share the work. They also need:

• Redis streams in multi-replica deployments. Set USE_REDIS_STREAMS=true. Without shared streams, saving a schedule fails rather than starting work another replica could not resume or stop.
• A durable checkpointer for agents that can pause. If the scheduled agent can use Ask User or needs tool approval, it needs Redis streams and a shared checkpointer even on a single process. The default MongoDB checkpointer works; type: memory does not. LibreChat rejects such a run before any model call when either is missing.
• The standard server entrypoint. The legacy experimental clustered entrypoint does not start the schedule engine.

Two environment variables control the engine (also listed in dotenv):