MCP server

@lokrix/mcp-server exposes the same GEO engine as Model Context Protocol tools — 200+ of them — so an agent can list projects, start a scan, read calibrated scores, and simulate optimization moves. It ships the lokrix-mcp CLI with two transports.

The Lokrix AI MCP server surfaceThree columns under the MCP server: tools (colour-coded as read-only or credit-spending), resources (project, scan and promptset URIs), and guided prompts such as audit-my-site and weekly-geo-review.MCP serverstdio + Streamable HTTPTools (lokrix.*)ResourcesPromptsanalyze_urlscore_promptstart_scanget_scanlist_scansget_visibilityget_prompt_setlist_competitorscompare_competitorsget_historylist_recommendationssimulate_changeget_optimization_planlist_projectscreate_reportproject://scan://promptset://audit-my-siteweekly-geo-reviewoptimize-for-promptexplain-scorefind-earned-media-gapspath-to-number-onereadspends credits

Two transports

  • stdio (the default) runs lokrix-mcp as a subprocess of an MCP client such as Claude Desktop — best for a single local agent.
  • Streamable HTTP (--http or MCP_HTTP_PORT) exposes the same tools over HTTP at /mcp — best for shared or remote agents.

Both read LOKRIX_API_URL and LOKRIX_API_KEY from the environment. The key’s scopes bound what the agent can do — exactly as they bound the REST API.

Connect an agent

Point any MCP client at the package over stdio:

{
  "mcpServers": {
    "lokrix": {
      "command": "npx",
      "args": ["-y", "@lokrix/mcp-server"],
      "env": {
        "LOKRIX_API_KEY": "sk_live_...",
        "LOKRIX_API_URL": "https://api.lokrix.de"
      }
    }
  }
}

Or run the HTTP transport for shared agents:

MCP_HTTP_PORT=3001 npx @lokrix/mcp-server --http
# → POST MCP requests to http://host:3001/mcp

Tool domains

Tools are grouped by resource — the same shape as the REST resources:

  • Projects — lokrix.list_projects · lokrix.create_project · lokrix.get_project · lokrix.update_project · lokrix.delete_project
  • Scans & engine-runs — lokrix.analyze_url · lokrix.start_scan · lokrix.get_scan · lokrix.list_scans · lokrix.cancel_scan · lokrix.estimate_scan_cost · lokrix.get_scan_advice
  • Scores & analytics — lokrix.score_prompt · lokrix.get_visibility · lokrix.get_history · lokrix.get_engine_breakdown · lokrix.get_analytics_overview
  • Prompts — lokrix.get_prompt_set · lokrix.generate_prompts · lokrix.activate_prompt_set · lokrix.lock_prompt_set · lokrix.add_prompt · lokrix.add_prompts_bulk
  • Recommendations & optimize — lokrix.list_recommendations · lokrix.generate_recommendations · lokrix.get_optimization_plan · lokrix.simulate_change
  • Competitors — lokrix.list_competitors · lokrix.compare_competitors · lokrix.compare_head_to_head · lokrix.get_competitor_trends
  • Reports — lokrix.create_report · lokrix.list_reports · lokrix.get_report · lokrix.share_report
  • Goals · alerts · schedules — lokrix.list_goals · lokrix.create_goal · lokrix.list_alerts · lokrix.create_alert · lokrix.get_schedule · lokrix.set_schedule
  • Integrations · billing · usage — lokrix.list_integrations · lokrix.get_usage · lokrix.get_subscription · lokrix.get_credit_balance · lokrix.list_api_keys · lokrix.list_webhooks

Tool metadata keeps agents safe

Every tool declares a scope, readOnly, destructive, and provenance, so a client can gate destructive or credit-spending calls behind confirmation. A typical read-only flow — lokrix.list_projects → lokrix.get_visibility — spends nothing.

Which tools spend credits

Only live engine runs (lokrix.start_scan / lokrix.analyze_url against real engines) meter credits. Predictive tools — lokrix.simulate_change and lokrix.get_optimization_plan — run on the calibrated model and cost 0 credits. Use lokrix.estimate_scan_cost to preview a run against your credit balance first.