Skip to main content
The ModernLoop MCP Server exposes read and write tools on the default endpoint. Tool availability is specific to your endpoint, OAuth scopes, and organization entitlements. Call tools/list after connecting for the authoritative tool list and input schema for your session. Two of the read tools are skill guides rather than data operations: list_modernloop_skills and load_modernloop_skill document the correct workflow, tool order, safety rules, and pitfalls for a scheduling domain. Load a guide before starting domain work — see Skill guides. Use the data model to find the right object and canonical ID, and MCP setup to troubleshoot missing tools or scopes. Call get_me first: its tool_access field predicts whether each public MCP tool can succeed with the token’s scopes. Endpoint variant and connection-level include_tools/exclude_tools filtering can further restrict visibility, while a listed tool can still return INSUFFICIENT_SCOPE; use tools/list for the active connection’s visible tools.

Tool reference

delete_interview_module, delete_interview_module_member, and cancel_scheduling_task are destructive. Read the current record and obtain explicit confirmation before calling them. create_scheduling_task requires an idempotency_key; reuse that key only when retrying the same intended create.

Skill guides

Skill guides are Markdown playbooks served over MCP. They describe the correct tool order, required inputs, safety rules, and common pitfalls for one scheduling domain, so an agent does not have to infer a workflow from tools/list.
  1. Call list_modernloop_skills with topic keywords in query and a brief telemetry.intent. Omit query to list every applicable guide. Add include_header: true for descriptions, related guides, and bundled reference paths.
  2. Call load_modernloop_skill with the exact skill_name from that response. Skill names are case-sensitive and are never inferred from topic words.
  3. Use header_only: true to preview a guide, or resource_path to load one bundled reference without reloading the whole guide.
Both tools require telemetry.intent: a brief English statement of what you are doing. Never put candidate or employee data, argument values, or secrets in it — ModernLoop does not retain the raw text, and it is not a place for sensitive input. The beta catalog covers modernloop/task-triage, modernloop/candidate-investigation, modernloop/scheduling-task-creation, and modernloop/interviewer-capacity.
A guide is only listed and loadable when every tool it documents is visible to your token, endpoint, and connection filters. On /mcp/readonly, guides covering task writes are absent. An unavailable or unknown skill name returns the same NOT_FOUND response, so a failed load is never evidence that a guide exists.

Workflow prompts

Copy one of these into a connected MCP client. Replace the bracketed text with your context.

Triage my scheduling tasks

Use get_me first, then call list_scheduling_tasks with my_tasks: true for your open scheduling tasks. Group them by urgency, status, and queue; call get_scheduling_task for the highest-priority items. Do not change anything.

Create a scheduling task for a candidate

Find candidate [name or email], then find their active application for [job]. Read the application and its job stages, identify the requested stage, and summarize the target application and stage before creating a scheduling task. After I confirm, call create_scheduling_task with a new idempotency key and report the result.

Investigate why a candidate is not scheduled

Find candidate [name or email], then inspect their applications, scheduling tasks, schedules, and interview events. Explain the current status and blockers using only the returned data. Do not make changes.

Review interviewer capacity

List users matching [team, name, or module], then inspect the relevant users and interview-module training statistics. Identify people who are paused or near their configured limits. Do not make changes.

Troubleshoot a ModernLoop issue

Search the help knowledge base for [issue or error message] with list_help_articles, read the best matches with get_help_article, and summarize the recommended steps with canonical article links. If nothing relevant is found, say so and ask one clarifying question.

Find stale tasks in a queue

Find the task queue named [queue name], then list its scheduling tasks that have not been updated since [date]. Summarize status, assignee, urgency, and the next useful follow-up. Do not make changes.

Further reading