LoginSign Up
← Back to Analytics
Weekly · Analytics · Week of July 6, 2026

Analytics tools spent the week wiring into AI agents over MCP, turning dashboards into things an assistant can drive.

mcpagent-driven-analyticsai-governancebehavioral-depthai-traffic-attributioncrawl-source-quality
Y
By Yahya Tür
LinkedIn ↗
Generated 1h agoDrawn from 10 products

The week in analytics

The single clearest direction in analytics this week was the Model Context Protocol moving from novelty to plumbing. Hex turned its agent into an MCP client that pulls live context from connected apps and tools, Apify shipped MCP connectors that let its Actors read and write inside authenticated apps like Notion, Slack, and GitHub, and Dovetail added MCP write tools for comments, folders, and tags on top of its research repository. The common thread is a shift in what these products are for: not just surfaces a human reads, but surfaces an agent operates. The repository, the dashboard, and the scraper are all being re-cast as tools an assistant can call.

The second arc is AI graduating from preview into governed, generally available product. Omni moved AI Hub to GA and gated its AI skills behind required access grants — the governance counterweight that lets a regulated buyer turn the feature on. Dovetail opened a Deep research mode for sustained, multi-step investigation rather than quick lookups. Even the products not chasing AI directly are sharpening analytical depth: Plausible broke out AI-assistant referrals from ChatGPT, Claude, Gemini, and Perplexity as a first-class channel, acknowledging that attribution itself is shifting toward AI traffic. The week rewarded products that paired new AI surface area with the access controls and depth to make it usable.

Leaders

Dovetail had the strongest week among the research-analytics tools, shipping a Deep research mode that lets users opt into longer, more thorough analysis from the same chat toolbar. It builds on a steady run of releases — running code and rendering charts inline, carrying full context across turns, and pulling workspace Docs into context — that collectively turn the chat surface from a qualitative summarizer into an analyst that reads, computes, and cites. New MCP write tools for comments, folders, and tags extend that to making the repository operable by external agents.

Hex is betting the analytics workflow becomes agent-driven, and this week's architectural move was becoming an MCP client: the Hex Agent can now connect to external apps and tools over MCP and use them as live context for analysis. Around that core it added agent web search, a model picker, presence inside Codex, and connectors for repos and Figma plus AWS IAM role support — integration and enterprise-auth plumbing that widens what the agent can ground on.

Apify is rebuilding its Actor platform as MCP-first agent infrastructure, and MCP connectors are the release that closes the biggest gap in that story. Actors could previously only touch the open web; they can now act inside logged-in apps like Notion, Slack, and GitHub, with credentials brokered through a proxy the Actor never sees. Paired with interactive OpenAPI docs and a one-time approval gate for full-permission Actors, Apify is making Actors callable, documented, and permissioned for agents that did not write them.

Plausible is competing on analytical depth rather than just simplicity. The standout was a dedicated AI Assistants channel that isolates referral traffic from ChatGPT, Claude, Gemini, and Perplexity instead of burying it under generic referrers. It lands alongside full-URL breakdowns in Page reports, strict-order funnels, and user-journey path tracing — a deliberate, consistent move into behavioral depth beyond simple counts.

Omni is racing to make AI a governed, GA-grade layer of the BI stack. AI Hub reached general availability this week and AI skills gained required access grants, turning the AI surface from an experiment into a supported, permissioned part of the platform. The high, steady cadence underneath — calculation pushdown, approximate aggregates, compute routing — is the performance plumbing that lets those AI features run cheaply at scale.

Wildcards

Axiom is the observability player drifting into analytics territory. Its Metrics GA brings hyper-cardinality metrics into the same store as logs and traces under one query layer, and exposes the whole surface to AI agents through MCP and a dedicated metrics skill. Completing the observability triad and making it agent-queryable mirrors exactly the MCP-first pattern the BI tools are chasing, from the telemetry side. Its recent Correlations release, stitching logs, traces, and metrics into one investigation, reinforces the move up the stack.

updown.io is the deliberate counterpoint to the AI rush. While the sector wires up agents, updown keeps patiently hardening a narrow uptime-and-pulse monitor: this week it swapped its Montreal node for a Toronto one for better IPv6 reliability and bundled pagination, pulse string validation, and notification fixes. No AI, no MCP — just methodical widening of the probe network. It is a useful reminder that not every analytics product needs an agent story to keep shipping.

Themes that compounded

  • MCP became standard plumbing this week, with Hex turning into an MCP client, Apify shipping MCP connectors, Dovetail adding MCP write tools, and Axiom exposing its query layer over MCP.
  • AI features are crossing from preview into governed GA, most clearly as Omni shipped AI Hub GA with access-grant controls rather than raw, ungoverned AI.
  • Analytics products are widening from passive reading into agent-operable action, re-casting repositories, scrapers, and dashboards as callable tools.
  • AI traffic is becoming a first-class attribution category, evidenced by Plausible breaking out a dedicated AI Assistants referral channel.
  • A crawl-source caveat persists: Zoho Analytics' tracked feed is its BI marketing blog rather than a release log, and the two Apache Superset rows (slugs apache-superset and superset) are the same product whose public feed is only Helm-chart packaging bumps, so neither reflects the real 6.x product work.

Watch this week

The near-term question is how far the agent-operable pattern spreads and whether governance keeps pace. Dovetail has positioned chat, not the page, as the primary surface, so expect Deep research mode to deepen and more of the repository to become MCP-writable. Hex's MCP-client move and presence in Codex suggest continued effort to be present across the agent stack rather than just inside Hex. Omni's pairing of AI GA with access grants is the template to watch — whether other BI tools ship AI behind comparable controls or push it raw. On the data-quality front, treat the Zoho Analytics blog feed and the Superset Helm-chart feed with skepticism: their absence from the leaders list is a crawl-source artifact, and the duplicate Superset row should be deduplicated upstream rather than read as two products shipping.