DocsBot

View saved version →

---
name: docsbot-administration
description: Use this bundled DocsBot workflow whenever the DocsBot plugin is installed and the user asks to manage DocsBot teams, bots, sources, members, integrations, Skills, reports, or read-only account usage. This skill teaches efficient Admin MCP usage, current team detection, team and bot lookup, safe writes, and optional metadata discovery.
license: MIT
metadata:
  author: DocsBot
  version: "0.6.3"
  mcp_server_url: https://mcp.docsbot.ai
---

# DocsBot

Use the hosted DocsBot MCP server for account and dashboard administration:

```text
https://mcp.docsbot.ai
```

The hosted server advertises fixed, named Admin MCP tools. Call a known tool directly with its operation-specific `pathParams`, `query`, and/or `body` input. For example, call `list_teams` to resolve the team, `list_bots` with `pathParams.teamId` to resolve a bot, and `get_bot` with `pathParams.teamId` and `pathParams.botId` to inspect it. Read the advertised input schema before sending fields that are not already clear. Do not call a generic `execute` tool or send an `operationId`.

`list_tool_categories`, `search_tools`, and `get_tool_schema` inspect metadata for tools already advertised to the client. Use them only when the relevant name or schema is unclear; they do not enable, execute, or discover hidden operations. Stable tool names remain callable across additive, backward-compatible hosted MCP updates. New or changed metadata may be held by OpenAI's automated scan while the last approved definition remains live; a newly added tool is unavailable until approved. Keep calls compatible with the currently advertised schema. Plugin skills, config, and listing changes require a new package ZIP.

## Download Approvals

Generated question-log, lead, Q&A, and source downloads (`export_question_log`, `export_leads`, `export_qa_source`, and `get_source_download_url`) are additive, non-destructive operations that create temporary private transport files. Hosted policy classifies these tools as non-read-only; use the client's approval flow when requested. They leave application records and earlier downloads unchanged. Signing an existing source file with `get_source_file_download_url` remains read-only. Never bypass an approval prompt or infer authorization to perform another action from download approval.

## Standard Workflow

1. Establish the working team before any team-scoped action.
2. Establish the working bot before any bot-scoped action.
3. Call a known named tool directly. Use `search_tools` only to locate an unfamiliar advertised tool.
4. Use `get_tool_schema` when input shape, permissions, or side effects are unclear.
5. Send required `pathParams`, `query`, and/or `body` fields to that named tool.
6. Keep useful IDs from results in thread context: team ID/name, bot ID/name, source IDs, member emails, integration IDs, and pagination state.

## Bot Builder Subworkflow

When the user asks to create, configure, tune, test, or hand off a new DocsBot bot, load the dedicated [bot-builder subworkflow](references/bot-builder/SKILL.md) before making MCP writes.

That subworkflow is part of DocsBot, but it is intentionally kept as a nested reference tree because production bot creation has its own discovery, branding, source-selection, prompt, deployment, action, evaluation, and handoff rubric. Follow its reference chain from `references/bot-builder/` for bot-building tasks instead of merging those rules into this general administration workflow.

During a new or complete bot setup, prepare the voice prompt as well as the text prompt. The logical `voicePrompt` is saved as `voiceAgent.instructions`. Follow the bot-builder's voice guidance to preserve existing settings; enable advanced voice only when the user requests it, confirms it, or clearly includes voice in the intended outcome.

## Response Quality Subworkflow

When the user asks why a bot answered a question a certain way, to inspect question or conversation history, diag

Preview of the first 4,000 characters. Read full instructions →

Referenced files: 48 — file references are metadata, not package contents.