dashies-plugin

<div align="center"> <a href="https://dashies.ai"> <picture> <source media="(prefers-color-scheme: dark)" srcset="assets/wordmark-dark.png"> <source media="(prefers-color-scheme: light)" srcset="assets/wordmark-light.png"> <img src="assets/wordmark-light.png" alt="Dashies" width="300"> </picture> </a> ### Your AI builds the dashboard. Dashies keeps it alive. Publish AI-built HTML dashboards to a shareable URL and keep them refreshing on a schedule,<br>with no AI in the loop after the first build.     [Website](https://dashies.ai) · [Docs](https://docs.dashies.ai) · [Quickstart](https://docs.dashies.ai/start/quickstart) · [MCP endpoint](https://mcp.dashies.ai/mcp) · [Issues](https://github.com/Dashies-ai/dashies-plugin/issues) <img src="assets/claude-code.webp" height="18" alt="">&nbsp; Claude Code &nbsp;&nbsp;&nbsp; <img src="assets/codex.webp" height="18" alt="">&nbsp; Codex &nbsp;&nbsp;&nbsp; <img src="assets/cursor.webp" height="18" alt="">&nbsp; Cursor </div> <a href="https://dashies.ai"><img src="assets/hero.png" alt="A Dashies dashboard served at nocturne.dashies.ai: five headline figures, a monthly demand chart and a channel breakdown" width="100%"></a> <p align="center">_{Built by an AI agent from a BigQuery warehouse and published through this plugin. Every number on it is the output of saved SQL that Dashies re-runs without the AI.}</p> ## What's in the box - **The `dashies` authoring skill.** The playbook your AI follows to build a dashboard that can re-run its own SQL: which connection, one read-only `SELECT` per dataset, what each column means, and how often to refresh. It loads automatically after install. - **The Dashies publish MCP server.** Tools for publishing, versioning, scheduling and inspecting dashboards, over OAuth. There is no API key to paste. - **A URL that stays alive.** Every dashboard lives in a workspace, at `https://<workspace>.dashies.ai/<slug>`: versioned, members-only, and rebuilt from your warehouse on the cadence you choose. ## Install ### Claude Code ```sh claude plugin marketplace add https://dashies.ai/marketplace.json claude plugin install dashies@dashies --scope user ``` `--scope user` installs Dashies for every project rather than binding it to the directory you are in, which is the difference between the `dashies` skill existing in your next repository and not. Run these in a shell, then `/reload-plugins` in Claude Code or start a new session. Inside Claude Code, `/plugin install dashies@dashies` takes no scope flag and asks where to install instead - the answer is **Install for you (user scope)**. ### Codex ```sh codex plugin marketplace add Dashies-ai/dashies-plugin codex plugin add dashies@dashies ``` If the marketplace clone fails on SSH host-key verification, use the HTTPS URL: `codex plugin marketplace add https://github.com/Dashies-ai/dashies-plugin.git`. ### Cursor Cursor 2.5+: search **Dashies** under Customize -> Plugins, or run `/add-plugin` in the editor. The marketplace listing is pending Cursor's review; until it lands, the one-click **Add to Cursor** button in the [install guide](https://docs.dashies.ai/guides/install) adds the connector, and `npx skills add Dashies-ai/dashies-plugin --global --agent cursor` adds the skill. <details> <summary><b>Several clients at once, or any other MCP client</b></summary> <br> **Every plugin-capable agent on your machine, in one command.** The vendor-neutral [`plugins`](https://www.npmjs.com/package/plugins) CLI detects Claude Code and Cursor and installs Dashies into each: ```sh npx plugins add Dashies-ai/dashies-plugin ``` Its Windows agent detection is still maturing, so prefer the per-tool commands above there. **On Codex, pick one route and never both**: when this installer detects Codex it registers `dashies@plugins-cli`, while the native commands above register `dashies@dashies`, so running both shows two install cards. `codex plugin list` tells you which one you have. **Any MCP client, without the skill.** Dashies is a plain remote MCP server at `https://mcp.dashies.ai/mcp`. Claude web, desktop and Cowork add it under Settings -> Connectors; Codex can add just the connector with `codex mcp add dashies --url https://mcp.dashies.ai/mcp`. A bare connector publishes fine, but the authoring skill, which is most of the product's behaviour, only ships with the plugin. </details> > [!TIP] > **No tokens, ever.** Sign-in is a one-time browser step (OAuth 2.1 + PKCE with dynamic client registration). In Claude Code, run `/mcp`, select **dashies** and choose **Authenticate**; other clients open the browser on the first Dashies tool call. The consent screen asks which workspace the tool may publish into, so sign up at [dashies.ai](https://dashies.ai) first: signing up creates your workspace. The token lives in your OS keychain and refreshes itself. ## Your first dashboard Ask, in plain language: ```text Build me a Dashies dashboard of last month's signups by plan, and refresh it daily. ``` Your AI reads the schema of your connected source, writes and validates the SQL, publishes a spec, and hands back a link: ```text https://<workspace>.dashies.ai/signups-by-plan ``` No warehouse yet? The built-in `self` connection (your own no-PII Dashies usage metrics) is always available, so the [quickstart](https://docs.dashies.ai/start/quickstart) works with no warehouse and no paid plan. ## How the refresh works The model runs once, at authoring time. After that, refresh is a server-side schedule: it re-runs the saved SQL against your source and rewrites only the numbers. The HTML never changes, and no model runs again. ```mermaid flowchart LR subgraph once["Authored once, with AI"] A["You describe<br/>the dashboard"] --> B["Your AI writes a spec:<br/>connection, SQL, columns,<br/>page, cadence"] B --> C["publish_dashboard"] end C --> D[("workspace.dashies.ai/slug")] subgraph loop["On schedule, without AI"] E["Hourly, daily,<br/>weekly or monthly"] --> F["Re-run the saved SQL<br/>against your source"] F --> G["Write fresh numbers<br/>into the dashboard"] end G --> D ``` - **Layout never drifts.** Refresh swaps data, not markup, so a dashboard cannot break visually between runs. - **Deterministic and cheap.** No model in the refresh path means no re-interpretation and no tokens per run. - **Checked before it goes live.** A spec publish compiles, validates and seeds the dashboard with real numbers from your connection first, so a structurally broken or silently wrong dashboard is refused rather than published. ## Good to know - **Scheduled refresh.** Hourly, daily, weekly or monthly, with an every-N multiplier and a timezone anchor. Or refresh on demand. - **Version history.** Every republish snapshots the previous body. Twenty autosaves are kept, plus up to thirty named versions, and you can roll back to any of them. - **Safe renames.** The old slug 301-redirects to the new one, so links you already shared keep working. - **Access-gated, always.** A dashboard opens only for the members of its workspace; anyone signed out is sent to sign in. There is no public dashboard. - **Sandboxed rendering.** Published dashboards run under a strict sandbox CSP, isolated from the rest of the origin. - **Warehouse connections.** Postgres, BigQuery, Snowflake, Amazon Redshift, Databricks, Microsoft SQL Server and Oracle Database, connected in the web app so credentials never pass through your AI. Which of them a refreshing dashboard can publish against is read out of the database as the publish is judged, so the publish refusal names the current set and this sentence names a reading: on 2026-09-18 it was BigQuery, Databricks, Oracle Database, Postgres, SQL Server and Snowflake. Redshift can be connected, explored and validated, but not published against. ## The MCP tools <details> <summary><b>Every tool, grouped by what it is for</b> (the first call opens the sign-in)</summary> <br> **Publish and manage** | Tool | What it does | |---|---| | `publish_dashboard` | Publish a dashboard from a YAML spec the server compiles, validates and seeds for you; `dry_run` checks the same spec without publishing anything. Returns the stable URL. | | `update_dashboard` | Change the name, tags or chart kind, or rename the slug. The old slug 301-redirects. | | `get_dashboard` | Read back a published file. | | `delete_dashboard` | Retire a dashboard. The URL stops resolving and the bytes are removed. | | `list_dashboards` | Your active dashboards, newest first, cursor-paginated. | **Versions** | Tool | What it does | |---|---| | `list_dashboard_versions` | The prior body snapshots of a dashboard, newest first: 20 autosaves plus up to 30 named versions. | | `get_dashboard_version` | Read one snapshot's body, to inspect or diff it before restoring. | | `restore_dashboard_version` | Roll back to a snapshot. The current body is snapshotted first, so the restore is normally reversible. | | `update_dashboard_version` | Name a snapshot ("Before redesign") or clear its label. | **Authoring** | Tool | What it does | |---|---| | `check_readiness` | What Dashies already knows about the account in one round trip: connections and their health, what is readable behind each, existing dashboards, and the single next step. | | `list_connections` | The warehouse connections you can build against: id, engine, status, last verified. Never returns secrets. | | `introspect_schema` | The columns and types a dashboard's SQL may read, with approximate row counts. | | `explore_data` | One read-only `SELECT` with a small, complete answer, for the questions that come before design. | | `validate_cube_sql` | Prove a finished `SELECT` survives the executor that will run it unattended on every refresh. | **Refresh and spec** | Tool | What it does | |---|---| | `get_refresh_status` | Schedule, next run, last run, recent history and errors. Read-only. | | `set_refresh_schedule` | Cadence (manual, hourly, daily, weekly, monthly), an every-N multiplier and a timezone anchor. | | `trigger_refresh` | Refresh now instead of waiting for the schedule (paid plans, rate-limited). | | `get_dashboard_spec` | The stored spec of a spec-backed dashboard, verbatim: the starting point of an edit. Read-only. | | `derive_dashboard_spec` | Draft a spec from an older hand-authored refreshable dashboard. Stores nothing. | | `get_source_config` | The stored refresh settings of a hand-authored dashboard. Read-only. | The live tool schemas at `https://mcp.dashies.ai/mcp` are the contract; the [tools reference](https://docs.dashies.ai/reference/mcp-tools) documents each one. </details> ## Links - Website: https://dashies.ai - Documentation: https://docs.dashies.ai - Install guide: https://docs.dashies.ai/guides/install - Marketplace manifest: https://dashies.ai/marketplace.json - MCP endpoint: https://mcp.dashies.ai/mcp ## License [MIT](LICENSE)