Skip to main content

​
ChatGPT Integration

There are two ways to use NexSpace from ChatGPT. The MCP connector (recommended) gives ChatGPT the full governed tool surface with OAuth — no API keys to paste. The legacy custom GPT Action remains available. Requires developer mode (Settings → Apps → Advanced settings → Developer mode; available on Pro, Plus, Business, Enterprise, and Education plans).
  1. In ChatGPT, add a new connector with the NexSpace MCP endpoint:
  1. Choose OAuth. ChatGPT discovers the NexSpace authorization server automatically and opens the NexSpace consent page.
  2. Log in, review the requested permissions (grouped, with Read/Write badges), and Allow access.
Every tool the connection is scoped for becomes available in conversations:
  • Reads (default read-only bundle): schedules, staffing, analytics, CRM, compliance — scoped to your role and organization.
  • Writes (explicit scopes like crm:write): ChatGPT asks you to confirm write tools client-side, and medium-plus-risk actions additionally create a NexSpace approval — admins are notified and the action executes only after a human approves. Check outcomes with get_approval_status, which requires general:read — that scope is in the default connector bundle, so a read-only connection already has it. If you are minting a credential with write scopes, request general:read alongside them so it can poll its own approvals.
  • Connected apps (apps:read / apps:write): your connected Slack / QuickBooks / HubSpot actions through the same governed connector.

​
Deep research & company knowledge

The NexSpace MCP server implements OpenAI’s deep-research contract: two read-only tools, search and fetch, over your organization’s knowledge base (policies, procedures, compliance documents, handbooks, training materials).
These two cover the knowledge base only. search queries the document index and returns matching documents; fetch returns one document’s full text by the id search handed back. They do not reach shifts, staff, payroll, or CRM records — use the workforce verbs and category tools for those — and they do not search the open web. Both are read-only tools in the general category, so both are gated on general:read.
  • Deep research: select the NexSpace connector as a source when starting a deep research run. ChatGPT will search and read your org’s documents alongside the web.
  • Company knowledge (Business/Enterprise): once the connector is added, ChatGPT can draw on your NexSpace knowledge base when answering everyday questions.
Results respect your role and organization: search only returns documents you can access, and fetch applies the same access conditions, so a document your search couldn’t surface can’t be fetched either. Both tools are covered by the default read-only connector bundle (general:read) — no extra scopes needed.

​
Legacy: Custom GPT Action

The original integration path — a custom GPT with REST Actions and an API key. Search “NexSpace Workforce Manager” in the GPT Store, or build your own:
  1. Go to ChatGPT → Create a GPT
  2. Actions → Create new action
  3. Set authentication: API Key → Bearer → paste your NexSpace API key
  4. Paste the OpenAPI schema from tools/chatgpt-actions/openapi-actions.yaml
  5. Copy instructions from tools/chatgpt-actions/gpt-instructions.md
Available REST actions include listFacilities, getFacility, searchStaff, getStaffMember, listShifts, and fillShift; the mcpToolCall passthrough reaches the MCP verbs (verifyCredential, previewPayrollRun, qualifyLead, …). Actions that modify data are marked x-openai-isConsequential: true, so ChatGPT asks for confirmation before executing. mcpToolCall is marked consequential too — it is a single operation that can reach any verb, mutating ones included, so it is confirmed even when the verb behind it is a read. Writes reached through the passthrough are additionally governed by the NexSpace approval gate: a medium-plus-risk verb returns {"status":"pending_approval","approvalId":…} instead of executing, and a human resolves it in NexSpace. ChatGPT’s confirmation is a client-side prompt, not the authorization decision.

​
Which should I use?

Use the MCP connector unless you specifically need a shareable custom GPT.