Beamfar Website Monitoring

Read-only MCP server for two-region website monitoring: check site status, open incidents, pre-incident DNS/SSL/SEO changes, security findings, and reports.

Размещённый MCP-сервер

npx add-mcp 'https://api.beamfar.com/mcp'

Устанавливается в Claude Code, Codex, Cursor и другие

Документация

What the Beamfar MCP server is

MCP, the Model Context Protocol, is the open standard AI assistants use to call tools on a service. The Beamfar MCP server lets an assistant read one organisation’s monitoring data: sites and status, incidents with two-region evidence, likely causes, changes, security findings and monthly reports. It only reads, and reports missing measurements as “data unavailable”.

FactValue
AddressPOST https://api.beamfar.com/mcp (a GET answers 405)
TransportMCP Streamable HTTP; JSON answers; no session; no server-sent stream
Protocol versions2025-11-25, 2025-06-18, 2025-03-26 and 2024-11-05, chosen in initialize
Sign-inOAuth 2.1 (PKCE S256 for Claude.ai, ChatGPT and local loopback CLIs) or Authorization: Bearer bfk_... read-only API key
ToolsTwelve, all marked read-only
PlansEvery plan, Free included, within its monthly request quota (pricing); write keys come with Agency and Agency Plus.
HostingServer and database in Falkenstein, Germany (Hetzner), in the EU
LimitsRequest up to 32 KiB, answer up to 256 KiB, tool call up to 25 seconds, two tool calls at a time per key; API rate limits apply

What you can ask

Ask in plain words and your assistant picks from these twelve read-only tools:

ToolAsk things likeWhat it reads
list_sitesWhich of my sites are down right now?Every site’s status, open incidents, last check and organisation counts
site_statusHow has shop.example.com done this week?One site: status, monitors, uptime and coverage over 1 to 2160 hours
list_incidentsWhat incidents did we have this month?Incidents newest first, for one site or all, open or resolved
incident_evidenceShow me the proof for that incident.Stored observations that opened and closed it: region, timings, HTTP, TLS and DNS facts
explain_incidentWhy did it go down?Ranked likely cause with confidence, fix steps, ruled-out causes and gaps
what_changedWhat changed on this site before it broke?DNS, certificate, redirect, SEO-signal, domain-expiry and deploy changes
first_day_findingsWhat should I look at first across all my sites?Ranked findings: expiring certificates and domains, redirects, SEO risks, DNS issues, sites down
monthly_report_summarySummarise last month’s report for this client.Stored monthly report: uptime, incidents, changes, expiring certificates
security_overviewWhich sites show signs of hacking or SEO spam?Scanned sites, coverage and per-site warning counts
site_securityWhat did the last security scan of this site find?Findings with kind, severity, confidence and evidence; observations, never a verdict
list_maintenanceIs any maintenance planned or running?Maintenance schedules and active or next windows
dns_changesDid anyone change a DNS record?Changes in watched NS, MX, SPF, DMARC, CAA and TXT records

Missing measurements return “data unavailable”; likely causes are ranked readings with confidence, never guesses.

Set-up for each assistant

Connect in two ways: sign in directly with OAuth 2.1 (Claude.ai, ChatGPT, Claude Code, Codex, Gemini CLI) or create a read-only key under Integrations & API -> AI assistants (more about keys) and press Test connection. Official add-ons for Claude Code, Gemini CLI and Codex live in github.com/Beamfar/integrations.

AssistantWhat we didDocumentation
Claude CodeTested on 3 November 2026: it connected and a Claude model called two tools; official plugin publishedClaude Code, MCP
Claude DesktopHelper tested, app set up per its docsLocal servers, mcp-remote
Claude.aiOAuth 2.1 callback enabled; connector set up per its own docsClaude connectors
Gemini CLITested on 3 November 2026: it reports the server as connected; official extension publishedGemini CLI, MCP
OpenAI CodexSet up per its own docs, not run by us; plugin, config.toml and skill publishedCodex add-on
ChatGPTOAuth 2.1 callback enabled; connector set up per its own docsChatGPT, MCP
CursorSet up per its own docs, not run by usCursor, MCP
VS Code and GitHub CopilotSet up per its own docs, not run by usVS Code, MCP configuration
WindsurfSet up per its own docs, not run by usWindsurf, customizations
ZedSet up per its own docs, not run by usZed, MCP
n8nSet up per its own docs, not run by usn8n, MCP Client Tool

Tested means we ran it against a test installation with made-up demo data. Per its own docs means we followed the assistant’s documentation. Never commit an API key. Product names belong to their owners and none endorses Beamfar.

Claude Code (Plugin & CLI)

claude plugin marketplace add beamfar/integrations
claude plugin install beamfar@beamfar

Or connect directly with OAuth 2.1 or a key: claude mcp add --transport http beamfar https://api.beamfar.com/mcp --header "Authorization: Bearer bfk_...". Add --scope user for every project.

Claude.ai & Claude Desktop

In Claude.ai or Claude Desktop, open Settings -> Connectors -> Add custom connector and paste https://api.beamfar.com/mcp to sign in with OAuth 2.1. For local config with an API key, put this in claude_desktop_config.json (pinned to mcp-remote@0.1.16, fixing CVE-2025-6514):

{
  "mcpServers": {
    "beamfar": {
      "command": "npx",
      "args": ["-y", "mcp-remote@0.1.16", "https://api.beamfar.com/mcp", "--header", "Authorization:${AUTH_HEADER}"],
      "env": { "AUTH_HEADER": "Bearer bfk_..." }
    }
  }
}

OpenAI Codex CLI & App

codex mcp add beamfar --url https://api.beamfar.com/mcp
codex mcp login beamfar

Or add [mcp_servers.beamfar] with url = "https://api.beamfar.com/mcp" and bearer_token_env_var = "BF_API_KEY" in ~/.codex/config.toml. The triage skill and Codex plugin live in github.com/Beamfar/integrations.

Gemini CLI (Official Extension & CLI)

gemini extensions install https://github.com/Beamfar/integrations

Installs the MCP server, GEMINI.md triage rules, and /beamfar:check-sites. Or add the endpoint directly: gemini mcp add --transport http --header "Authorization: Bearer bfk_..." beamfar https://api.beamfar.com/mcp. Tested with Gemini CLI 0.66.0.

ChatGPT Connectors

Open Settings -> Connectors -> Create custom connector, enter https://api.beamfar.com/mcp, and sign in with OAuth 2.1. Revoke access anytime under Settings -> Connected assistants in the app.

Cursor

Or save in ~/.cursor/mcp.json with BF_API_KEY in your environment:

{
  "mcpServers": {
    "beamfar": {
      "url": "https://api.beamfar.com/mcp",
      "headers": { "Authorization": "Bearer ${env:BF_API_KEY}" }
    }
  }
}

VS Code and GitHub Copilot

Or save in .vscode/mcp.json (VS Code prompts for the key and keeps it out of the file):

{
  "inputs": [
    { "type": "promptString", "id": "beamfar-key", "description": "Beamfar API key", "password": true }
  ],
  "servers": {
    "beamfar": {
      "type": "http",
      "url": "https://api.beamfar.com/mcp",
      "headers": { "Authorization": "Bearer ${input:beamfar-key}" }
    }
  }
}

Windsurf

Save in git-ignored .devin/mcp_config.local.json:

{
  "mcpServers": {
    "beamfar": {
      "url": "https://api.beamfar.com/mcp",
      "transport": "http",
      "headers": { "Authorization": "Bearer bfk_..." }
    }
  }
}

Zed

Add to Zed settings (with OAuth 2.1 or an Authorization header):

{
  "context_servers": {
    "beamfar": {
      "url": "https://api.beamfar.com/mcp",
      "headers": { "Authorization": "Bearer bfk_..." }
    }
  }
}

n8n

Add the MCP Client Tool node to an AI Agent with endpoint https://api.beamfar.com/mcp, transport HTTP Streamable, and Bearer auth in a credential.

Any other assistant

Any Streamable HTTP MCP client works with https://api.beamfar.com/mcp using OAuth 2.1 or Authorization: Bearer bfk_.... Also published as com.beamfar/website-monitoring in the Official MCP Registry. See the help article and API documentation.

What it does not do

  • It does not write. All twelve tools are read-only; billing, members and settings are out of reach.
  • It does not push. The server keeps no session; alerts arrive through your configured channels.
  • It has no resources or prompts on the server itself — slash commands and triage skills ship in our assistant add-ons.

Keys and safety

  • Every OAuth grant and API key is scoped read-only to one organisation and acts as the member who created it. Revoking in Settings stops access at once.
  • Names, URLs, titles and DNS records from monitored sites are marked untrusted data so monitored text cannot steer your assistant.
  • Browser Origin headers are refused unless allow-listed; we count calls per key and never store arguments or answers.
  • Questions and tool answers pass through the company that runs your assistant under its terms. See prompt injection defence and security.

How we tested it

On 3 November 2026 we ran a test installation with demo data (12 sites, two regions, 30 days of checks):

  • Official MCP Python SDK 1.30.0 ran initialize, tools/list and four tool calls; MCP Inspector CLI version 2.13.0 verified all twelve schemas.
  • mcp-remote version 0.1.16 connected and answered what_changed.
  • Claude Code 2.1.316 connected and called list_sites and site_status; Gemini CLI 0.66.0 connected and reported ready.
  • Live https://api.beamfar.com/mcp answers 401 with WWW-Authenticate: Bearer and OAuth metadata discovery, and 405 on GET.

We have not run Cursor, VS Code, Windsurf, Zed, n8n or the Claude Desktop app. Report any issue via contact.

Questions

What is the Beamfar MCP server?

A hosted, read-only server letting AI assistants read one organisation’s sites, incidents, evidence, changes and monthly reports through twelve tools via OAuth 2.1 or an API key.

Can the assistant change or delete anything in my account?

No. All twelve tools only read; billing, monitors, alerts and members cannot be modified.

Which plans include the MCP server?

Every plan, Free included (3,000 requests a month on Free, more on paid plans). Keys that write come with Agency and Agency Plus.

Does it work with Claude.ai, ChatGPT, Codex and Gemini CLI?

Yes. MCP over OAuth 2.1 (PKCE S256) supports Claude.ai, ChatGPT and local loopback CLIs (Claude Code, Codex, Gemini CLI), alongside read-only API keys and our official add-ons on GitHub.

Where is the data processed?

The server and database run in Falkenstein, Germany, in the EU. Tool answers pass through your assistant provider under its terms.

What if a key leaks or I want to disconnect an assistant?

Revoke the API key or OAuth grant anytime in Settings: access stops immediately.