Acumen MCP server

Who it's for: Engineers and operators who want to ask an AI assistant (Cursor, Claude, ChatGPT-style agents) about their monitoring data in plain language.
You'll learn: How to connect an MCP client to Acumen Logs, which tools the assistant can use, and how to troubleshoot the connection.

The Acumen MCP server exposes your monitoring data through the Model Context Protocol. Once connected, an assistant can answer questions like "Is anything failing right now?" or "Why did the checkout journey fail last night?" by reading your projects, monitors, check history and status pages.

It is a hosted server and uses the same API keys, permissions and data as the Acumen API. Read tools work with any key. Write tools, which change uptime, synthetic, SSL and WHOIS monitors, their alerts, and account notification channels, require a key created with Allow changes.


Prerequisites

  • An Acumen Logs account on a paid plan.
  • An API key from Account > API Keys. See Create an API key. We recommend a dedicated key per assistant, for example Cursor MCP.
  • An MCP client that supports remote (Streamable HTTP) servers, such as Cursor, Claude Code or Claude Desktop with mcp-remote.

Server details

Item Value
URL https://app.acumenlogs.com/mcp
Transport Streamable HTTP (POST)
Authentication Authorization: Bearer <your-api-key> header
Access Read tools for every key. Write tools only when the key has the write ability. Same project and team permissions as the dashboard
Rate limit 120 requests per minute per API key
Timestamps UTC

The API Keys page in your account shows the server URL and ready-to-copy configuration for each client. Right after you generate a key, the snippets on that page include it automatically.


Connect your client

Replace <your-api-key> with the key you generated.

Cursor

Add the server to ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project):

{
  "mcpServers": {
    "acumen": {
      "url": "https://app.acumenlogs.com/mcp",
      "headers": {
        "Authorization": "Bearer <your-api-key>"
      }
    }
  }
}

Open Cursor Settings > MCP and check that acumen shows a green status with its tools listed.

Claude Code

claude mcp add --transport http acumen https://app.acumenlogs.com/mcp \
  --header "Authorization: Bearer <your-api-key>"

Run claude mcp list to confirm the connection.

Claude Desktop and other stdio-only clients

Clients that only launch local servers can connect through the mcp-remote bridge (requires Node.js). Add this to claude_desktop_config.json:

{
  "mcpServers": {
    "acumen": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://app.acumenlogs.com/mcp",
        "--header",
        "Authorization:${AUTH_HEADER}"
      ],
      "env": {
        "AUTH_HEADER": "Bearer <your-api-key>"
      }
    }
  }
}

The header is passed through an environment variable because some clients split arguments on spaces. Restart Claude Desktop after saving.

Any other MCP client

Point the client at https://app.acumenlogs.com/mcp using the Streamable HTTP transport and send the Authorization: Bearer <your-api-key> header on every request.


Tools

Read tools are idempotent. List tools accept page (default 1) and per_page (1–100, default 25) and return data plus a meta object with current_page, per_page, total, last_page and has_more_pages.

Write tools (create-uptime, update-uptime, and the same pair for synthetic, ssl and whois, plus update-monitor-alerts and the notification-channel create and update tools) fail with an error when the key is read-only. Nothing can be deleted through the server. update-monitor-alerts sets alert channels on uptime, synthetic, API monitors, SSL, WHOIS and heartbeats. list-alert-channels returns the connected channel ids to pass in. They follow the same admin permission and plan limits as the dashboard. Connecting and disconnecting Slack stays in Account → Notifications.

Tools that take a type accept one of uptime, synthetic, api-monitors, ssl, whois or heartbeats.

Tool Arguments Returns
get-health-summary none Total, active, inactive, failing and healthy counts per monitor type, plus up to 20 monitors that need attention
get-account none Your user, timezone, plan, and the API key in use
list-projects page, per_page Projects (sites) with monitor counts per type
get-project id A single project
list-monitors type, optional site_id, active, page, per_page Monitors of one type with configuration and current state
get-monitor type, id One monitor's configuration and current state
get-check-history type, id, optional from, to, status (pass or fail), page, per_page Check history, newest first
list-status-pages page, per_page Status pages and the monitors shown on them
get-status-page id A single status page
list-announcements page_id, optional status, page, per_page Status page announcements, newest first

In get-health-summary, these count as "needs attention":

Type Condition
Uptime, synthetic, API monitors Last check failed
Heartbeats The next expected pulse is overdue
SSL Certificate expires within 14 days
WHOIS Domain expires within 30 days

The fields returned match the Acumen API response fields. Credentials, custom headers, request bodies, journey steps and heartbeat pulse URLs are never returned.


Prompts

Prompt Arguments What it does
incident-triage type, id, optional days (default 7) Guides the assistant through fetching a monitor and its failures, then summarising when they started, patterns by region, status code or time, and likely causes

In clients that support prompts (for example, as a slash command), pick incident-triage and fill in the monitor type and id.


Example questions

  • "Using Acumen, is anything failing right now?"
  • "Which SSL certificates or domains expire in the next month?"
  • "List the synthetic tests in the Checkout project and tell me which ones failed today."
  • "Show uptime monitor 123's failed checks since Monday. Are failures tied to one region?"
  • "Compare response times for the API monitors in project 42 over the last week."
  • "What did we post on our status page during the last incident?"

Tip: mention Acumen in the question so the assistant knows to use these tools.


Security

  • Write is opt-in. New keys are read-only unless Allow changes is selected. A read-only key can still connect; write tools reject it.
  • Same permissions as the dashboard. Team members only see projects and monitors they can already view.
  • One key per assistant. Revoke a key from Account > API Keys to disconnect that assistant immediately. The Last used column shows whether a key is active.
  • Prefer expiring keys. Use 30 or 90 days, and rotate keys on a schedule.
  • Keep keys out of git. Use the global ~/.cursor/mcp.json or environment variables rather than committing project config that contains a key.
  • Mind your AI provider. Data returned by tools is sent to the model your client uses. Review your provider's data handling policy before connecting production accounts.

Troubleshooting

Symptom Likely cause What to do
Client shows the server as failed or 401 Key missing, mistyped, revoked or expired, or the header is not Bearer <key> Generate a new key and check the header format in your config
402 Payment Required Free plan or cancelled subscription Upgrade at Subscription, then reconnect
403 Token without the read ability (for example a mobile app token) Create a key from the API Keys page
Tool returns "not found" for a monitor you can see Wrong type or id, or your team role lacks access Use list-monitors to confirm the type and id
Assistant doesn't use the tools The client hasn't loaded the server, or the question is ambiguous Check the MCP settings in your client, and mention Acumen in your question
429 More than 120 requests per minute for this key Wait a minute, or ask the assistant for narrower queries (filter by site_id, dates or status)
mcp-remote header errors Spaces in arguments are split by the client Use the AUTH_HEADER environment variable pattern shown above

Need help?

Contact support@acumenlogs.com with your client name and version, the key name (never the secret), and the error your client shows.