LoginSign Up
← Back to all sparks
C

Count

ANALYTICS
Velocity6.3

Analytics canvas tool combining SQL, Python, and visual analytics

count.co ↗

Count is turning its BI canvas into a governed, agent-operated analytics platform.

agentic-analyticsmcppublic-apiwarehouse-connectorssemantic-modelsdata-canvas
◆Current state
Count is a data-canvas analytics tool reorganizing itself around an AI agent. In two months it shipped a full public REST API and hosted MCP server (governed agent access via OAuth and service accounts), a major agent upgrade that lets the agent read and edit the entire canvas and answer from Slack, and the ability to plug external MCP servers (Linear, HubSpot, Stripe, Slack, Drive) into the agent. Around the agent it keeps broadening warehouse support—ClickHouse, Snowflake semantic models, OSI—alongside chart and UX polish.
◆Where it's heading
Count is building toward analytics where agents are first-class operators: a governed API/MCP layer for access, an agent that drives the canvas end to end, external tool reach via MCP, and connection-level context so guidance is captured once and inherited. Governance—permissions, scopes, service accounts—is the enabling layer that makes agent access acceptable in real data stacks rather than a bolt-on.
◆Prediction
Expect more connection- and warehouse-level context controls, a widening catalog of supported external MCP integrations, and deeper Slack-native agent workflows.

◆Recent moves

  1. 15d ago

    Connect external MCP servers to the Count agent

    ⚡ SPARK

    This is the release where Count's agent stops being warehouse-bound: it can now call external MCP servers to pull issues from Linear, enrich customers from HubSpot or Stripe, or search Slack—and connection-level context lets a data source's conventions be captured once and inherited by every agent working against it.

  2. 29d ago

    Dashed lines

    A feature-and-fix release: dash styles as both a direct line property and a visual mapping, built-in country/US/EU/UK boundary data for instant maps, and a resume-agent tool over MCP to continue an existing Count agent conversation from a connected client.

  3. 1mo ago

    New workspace home

    A redesigned workspace home, table totals and sub-totals, and inspectable DuckDB query profiles—orientation, summarization, and performance-visibility improvements to the canvas experience.

  4. 1mo ago

    ClickHouse support

    Native ClickHouse connectivity (including over SSH tunnels) extends Count's warehouse reach, and a public API for canvases lets cells be read and executed programmatically so canvases become building blocks for external apps and pipelines.

  5. 2mo ago

    Major Count agent upgrade: edits any cell, runs in Slack

    ⚡ SPARK

    The agent graduates from assistant to operator: it reads the canvas on its own, edits any cell, text, or frame (not just its own), checks that its edits actually run, searches the docs, and now answers directly in Slack—with faster browser data transfer underneath.

  6. 2mo ago

    Public API and MCP server

    ⚡ SPARK

    This is the platform foundation behind everything since: a full public REST API (OpenAPI, service accounts) and a hosted MCP server that acts as a governed, permission-inheriting front door for agents into the workspace.