Overview
The TrustData MCP server lets any MCP-compatible AI agent (Claude Desktop, Claude Code, Cursor, Gemini CLI, ChatGPT) query your marketing data, review past experiments, create recommendations, and trigger GEO probes.
It's the same surface as the public REST API (/api/v1/), exposed over the Model Context Protocol. Your agent picks the right tool for each step; you don't write glue code.
Who this is for
- Marketing operators who want to chat with their dashboard ("why did ROAS drop yesterday?") instead of clicking through it.
- SEO × GEO specialists driving a weekly loop of diagnose → propose → ship → measure, with the agent doing the diagnosis and bookkeeping.
- Agencies who manage many clients and want an LLM to triage alerts and draft recommendations per property.
- Developers automating reports, anomaly digests, or connector QA without maintaining a scraping layer.
What you get
26 tools across five families:
Bootstrap / discovery
list_properties, list_data_sources, list_attribution_ids, list_brand_prompts, list_competitors, list_analytics_queries: enumerate the UUIDs and metadata every other tool needs.Analytics primitives
query_metrics, get_timeline, get_attribution, get_geo, get_retention, get_ltv, list_change_events: the same numbers the dashboard shows, queryable by date range and dimension.AI Visibility (GEO) & SEO
get_geo_visibility, list_seo_keywords, list_citation_gaps, run_probe: see where your brand ranks in LLM answers, find publishers citing you without a link, probe prompts on demand.Recommendations & experiments
list_recommendations, create_recommendation, mark_recommendation_done, dismiss_recommendation, list_concluded_experiments: the agent proposes hypotheses, tracks outcomes, learns from prior winners.Alerts & anomalies
list_alerts, list_anomalies, acknowledge_alert, resolve_alert: triage queue for when something in paid / SEO / GEO goes sideways.How it works
- Scopes = capabilities. A read-only token can never write, even if the LLM asks it to.
- Tier = rate limit. Measure (30/min) → Optimize (300/min) → Prove (3000/min). Agentic workflows burn through Measure quickly; use Prove.
run_probeis separately metered against your GEO add-on's monthly engine-call allowance.
Pricing
The MCP endpoint is included with every paid TrustData plan. The only per-call cost is run_probe, which consumes engine calls from your GEO add-on. Read-only tools (everything else) are flat-rate and rate-limited, not metered.
Next step
Issue a token and connect your agent in 2 minutes → Quickstart