generated: '2026-08-06'
method: derived
source: mcp/asknicely-mcp.yml + openapi/asknicely-openapi.yml
description: >-
  Binds AskNicely's MCP capability surface (Ask NiceAI) to the REST operations in
  openapi/asknicely-openapi.yml. AskNicely's live tools/list is OAuth-gated and it publishes no static
  tool manifest, so no verified tool NAMES exist to map. Every row below is therefore a CAPABILITY AREA
  that AskNicely names in its own Ask NiceAI documentation, matched by semantics to the REST operation
  that serves the same data — confidence is `low` throughout by construction, and no row should be read
  as a claim about a tool's real name or inputSchema. Resolving this requires an authenticated
  tools/list against a tenant with the Ask NiceAI add-on.
surfaces:
  openapi:
    file: openapi/asknicely-openapi.yml
    operations: 12
    gated: false
  mcp:
    url: https://{domain}.asknice.ly/mcp
    probed: https://nicely.asknice.ly/mcp
    tools_list_gated: true
    gate: OAuth 2.1 bearer token (scope `tools`) + Ask NiceAI add-on entitlement
  graphql:
    present: false
crosswalk:
- tool: NPS summaries
  category: analytics
  rest: [getNps]
  binding: rest
  confidence: low
  note: >-
    Ask NiceAI answers natural-language NPS questions; /getnps/{days} is the REST operation returning the
    same score, with the same filters[]/values[] custom-field filtering.
- tool: Survey responses
  category: feedback
  rest: [getResponses]
  binding: rest
  confidence: low
  note: >-
    Ask NiceAI cites the specific customer comments behind an answer; /responses is the REST operation
    returning those response rows, including comment, score and question_type.
- tool: Recipient lookup ("who was sent this survey", "who hasn't responded yet")
  category: feedback
  rest: [getResponses, getContact]
  binding: rest
  confidence: low
  note: >-
    Fans out. The `raw` filter on /responses returns surveys sent but not responded to, which is the REST
    equivalent of the "who hasn't responded" question; /contact/get resolves an individual contact.
- tool: Leaderboards / frontline performance
  category: analytics
  rest: [getSentStats, getHistoricalStats]
  binding: rest
  confidence: low
  note: >-
    Partial. The REST statistics operations return account- and segment-level aggregates; AskNicely's
    per-person and per-location leaderboards have no REST equivalent.
mcp_only:
- tool: Reports and visualizations (HTML artifacts, saved reports)
  reason: >-
    Ask NiceAI generates charts, heatmaps and interactive HTML artifacts and can save/re-run them as
    Saved Reports. There is no REST operation that produces or stores a report artifact.
- tool: Help-center / product knowledge search
  reason: >-
    Ask NiceAI searches AskNicely's own knowledge base. Not a REST capability; the underlying corpus is
    the Zendesk help centre, not the customer's account data.
- tool: Themes / AI insights
  reason: >-
    AI Themes, AI Insights and Moderation are AI-layer products with no public REST surface.
- tool: Case management and workflow configuration questions
  reason: >-
    Ask NiceAI can answer on case status, assignees and workflow execution stats. The REST API exposes
    `status` and `case_closed_time` on a response row but has no case or workflow operations.
- tool: Related Zendesk ticket lookup
  reason: Cross-system join into Zendesk performed server-side; no REST operation exposes it.
rest_only:
- capability: Contact import and survey triggering
  operations: [triggerSurvey, bulkAddContacts, uploadContactsCsv]
  reason: >-
    Ask NiceAI is documented as read/analysis over account data. No documented MCP capability writes
    contacts or sends surveys.
- capability: Contact lifecycle
  operations: [removeContact, deactivateAllContacts, getUnsubscribedContacts]
  reason: No documented MCP capability deactivates contacts or reads the unsubscribe list.
- capability: Privacy / GDPR erasure
  operations: [privacyRemoveContacts]
  reason: >-
    Destructive and irreversible (the contact is blocklisted). No documented MCP capability performs it,
    and it should not be exposed to an agent without human approval.
coverage:
  tools_named: 0
  capability_areas_documented: 10
  capability_areas_bound_to_rest: 4
  mcp_only: 5
  rest_operations_total: 12
  rest_operations_with_a_capability_match: 6
  confidence_ceiling: low
gaps:
- Tool names and inputSchemas require an authenticated tools/list against a tenant with the Ask NiceAI add-on.
- AskNicely publishes no llms.txt and no static MCP tool manifest, so no name-level fallback exists.