Skip to content
Nimbly Technologies logo
Book a Demo
Model Context Protocol

Ask your AI assistant about your frontline operations.

The Nimbly MCP server connects Claude, ChatGPT and any other MCP-compatible assistant to your Nimbly account. Sites, audit reports, issues, users, questionnaires, schedules and departments — read-only, scoped to your organization, no data export required.

  • 12 tools
  • Read-only
  • OAuth 2.1 + PKCE
  • Streamable HTTP
  • Single-organization scope
MCP server endpoint https://api-public-v2-bgd5wvtq7a-df.a.run.app/mcp
What it does

Your operations data, in the conversation

Nimbly already records what happens across your stores, branches and plants. The MCP server lets an assistant read that record directly, so questions get answered from your live data instead of a stale export.

Audits and compliance

Pull submitted reports with their scores, questions, answers and photos over any date range, and summarise compliance by site, department or questionnaire.

Issues and corrective actions

Filter findings by status, severity, priority, site, assignee or due date, and read the full activity trail of any single issue to see how long it sat where.

Sites, people and plans

Resolve site, user, department and questionnaire names into records, and check audit schedules — planned, in progress, completed or overdue — across a date range.

Example prompts

Things people ask on day one

  • “Show me the five sites with the most open issues this month.”
  • “List the audit reports submitted last week and summarise compliance rates.”
  • “Which auditors completed the most reports in July?”
  • “How long did our critical issues in Jakarta take to close last quarter?”
  • “Which scheduled audits were missed at facility sites in the last 30 days?”
  • “Give me the full detail of issue X, including its status history and photos.”
Capabilities

12 tools across seven data domains

Every tool is read-only and carries its own OAuth scope, so you grant exactly the access you want. The usual pattern is a list_ call to find an identifier, then a get_ call for full detail.

read:sites

Sites

  • list_sites — stores, branches and outlets with their departments, city and timezone. Filter by site name, department, city or location.
  • get_site — one site in full: address, timezone, departments and assigned users.
read:reports

Audit reports

  • list_reports — submitted reports with scores and completion metadata. Filter by site, department, questionnaire, submitter, city, status and date range.
  • get_report — one report in full: every question, answer, score and attached photo.
read:issues

Issues

  • list_issues — findings and corrective actions. Filter by status, severity, priority, origin, site, department, category, creator, assignee, city and date range, with sorting.
  • get_issue — one issue in full: status, severity, priority, assignees, due date, linked report and photos.
  • get_issue_status_history — the chronological activity trail, with real status transitions separated from comments and reassignments.
read:users

Users

  • list_users — active users with role, email and status, searchable by name or email.
  • get_user — one user: name, email, role, status, language and onboarding state.
read:questionnaires

Questionnaires

  • list_questionnaires — audit checklists in your organization, most recently updated first, with the question count of the latest version.
read:schedules

Schedules

  • list_schedules — the audit plan over a date range, each schedule with its date and status (not started, in progress, completed or overdue), plus per-status counts.
read:departments

Departments

  • list_departments — departments defined in your organization, including disabled ones, with their status.

Built to answer accurately

Responses are shaped for assistants, not just for programs: the date window actually applied comes back with the results, partial pages say so, and filters backed by a fixed set of values are declared as such — so an assistant cannot quietly answer “none” when it means “not that filter”.

Getting connected

Add Nimbly to your assistant

The server is a remote MCP server over Streamable HTTP, so there is nothing to install locally. Point your assistant at the endpoint and sign in to Nimbly — the connection is authorised with OAuth and limited to the scopes you approve.

Claude (web, desktop and mobile)

  1. Open Settings → Connectors.
  2. Choose Add custom connector.
  3. Paste the Nimbly MCP server URL.
  4. Sign in with your Nimbly account and approve the scopes.

Custom connectors are available on Claude Pro, Max, Team and Enterprise plans. Our listing in the Claude Connectors Directory is in review; the steps above work today either way.

ChatGPT

  1. Open Settings → Connectors (or Apps & Connectors).
  2. Add a connector using the Nimbly MCP server URL.
  3. Complete the OAuth sign-in to your Nimbly account.

Connector support depends on your ChatGPT plan and workspace settings; on some plans it lives behind developer mode. Our submission to the OpenAI app directory is in progress.

Claude Code and other MCP clients

claude mcp add --transport http nimbly \
  https://api-public-v2-bgd5wvtq7a-df.a.run.app/mcp

Prefer a service credential over an interactive login? Any client can send a Nimbly organization API key instead:

claude mcp add --transport http nimbly \
  https://api-public-v2-bgd5wvtq7a-df.a.run.app/mcp \
  --header "x-api-key: <your-api-key>"

Generic client configuration

{
  "mcpServers": {
    "nimbly": {
      "type": "http",
      "url": "https://api-public-v2-bgd5wvtq7a-df.a.run.app/mcp"
    }
  }
}

Need an API key, or not sure whether your plan includes API access? Ask your Nimbly customer success manager, or talk to us.

Security and data handling

Read-only, single-organization, nothing stored

The MCP server is the same public API your integrations already use, exposed through a protocol assistants understand. It adds no new write paths and no new data stores.

No writes, ever

All 12 tools are reads. Nothing in the MCP surface creates, edits or deletes a site, report, issue, user or schedule.

One organization per connection

Every request is bound to the organization that authenticated it. No tool can reach another organization's data, whatever it is asked.

Scoped consent

Seven scopes, one per data domain. A token missing a scope is refused for the tools that need it, so you can grant issues without granting users.

Modern OAuth

OAuth 2.1 with PKCE (S256), dynamic client registration, short-lived access tokens, refresh token rotation and revocation.

No copy of your data

The server holds no cache or copy of your records. Each tool call reads live from the platform and returns the result to your assistant.

Fair-use limits

Requests are rate limited per organization on the same budget as the REST API — one tool call costs one request. Tool discovery is not metered.

Technical details

For the team reviewing this

Server URLhttps://api-public-v2-bgd5wvtq7a-df.a.run.app/mcp
TransportStreamable HTTP (POST /mcp), stateless — no session header required
Protocol revisionsMCP 2026-07-28 (handshake-free) and the 2025-era initialize flow, served by the same tool catalog
AuthenticationOAuth 2.1 (PKCE S256, dynamic client registration, refresh rotation, revocation), or a Nimbly organization API key via x-api-key or Authorization: Bearer
Protected resource metadata/.well-known/oauth-protected-resource (RFC 9728), advertised on 401 via WWW-Authenticate
Scopesread:sites read:reports read:issues read:users read:questionnaires read:schedules read:departments
Tools12, all annotated read-only, non-destructive and idempotent
Rate limitingPer organization, default 10 requests per second with a burst bucket; one tool call = one request. initialize and tools/list are unmetered
DatesYYYY-MM-DD. Issue and report listings default to the last 7 days and report the window they applied
FAQ

Common questions

Can the assistant change anything in Nimbly?

No. The MCP server exposes reads only. Creating an issue, editing a report or reassigning work still happens in the Nimbly app.

Who can see our data?

Only the assistant you connect, on behalf of the account that signed in, for the organization that account belongs to. Access ends when you revoke the connection in your assistant or the token is revoked on our side.

Do we need a paid Nimbly plan?

The MCP server runs on Nimbly's public API. If your plan includes API access you can connect today; if you are not sure, ask your customer success manager.

Is it available in the Claude and ChatGPT directories?

Directory listings are in review with both Anthropic and OpenAI. Until they are live you can add Nimbly as a custom connector with the same URL and get the same tools.

Which assistants work with it?

Any MCP client that speaks Streamable HTTP — Claude, Claude Code, ChatGPT connectors, and the growing set of third-party clients and IDE agents that support remote MCP servers.

What about data residency and retention?

The MCP server stores nothing. Records are read live from the Nimbly platform under your existing agreement; see our privacy policy and service level agreement.

Put your operations data one question away

Connect Nimbly to your assistant in a couple of minutes, or talk to us about rolling it out across your team.

Security & Privacy
ISO 27001 certifiedSOC 2 compliance badgeSecurity certification badge
LinkedInWhatsAppLINE