Validate Dashboardbase widget endpoint responses and setup files against the live contract.
The com.dashboardbase/mcp Model Context Protocol (MCP) server validates Dashboardbase widget endpoint responses and associated setup files against the live contract. Its purpose is to check that responses and configuration artifacts conform to the current expected specifications.
๐ ๏ธ Key Features
Validates Dashboardbase widget endpoint responses
Validates setup files against the live contract
๐ Use Cases
Contract validation for Dashboardbase widget endpoints
Verifying setup/configuration files meet the live contract requirements
Stop guessing whether your dashboard JSON is right. Ask.
This is the official MCP server for dashboardbase. Add it to your AI tool and it can check a widget endpoint's response โ or a whole dashboard setup file โ against the contract dashboardbase actually enforces, before you ever open the app.
The dashboardbase skill teaches your agent the JSON contract. This MCP lets it check its own work. Use both: the skill gets the shape right, the MCP proves it.
What this MCP does
It gives your agent two tools:
Tool
What it checks
validate_setup_file
A dashboardbase setup file โ the JSON that provisions a whole dashboard's widgets and datasources. Reports errors and warnings with the field, line and column.
validate_widget_response
The JSON body a widget endpoint returns, against the widget contract. Reports the path and message for each problem.
Both call the public dashboardbase validation API. No account, no API key, nothing to configure.
Before / after
Without it โ your agent writes an endpoint, you deploy it, you wire it into dashboardbase, the widget shows an error, you go read the docs, you fix it, you deploy again.
With it โ your agent writes the endpoint, validates the response, fixes the two things that were wrong, and hands you something that renders the first time.
code
Invalid setup file โ 2 errors, 1 warning
mappings[0].type 14:9 Unknown widget type "guage"
mappings[1].path 22:5 Path must start with "/"
warning mappings[2] No refreshInterval set
Install
Claude Code
bash
claude mcp add dashboardbase -- npx -y @dashboardbase/mcp
One click
Claude Desktop
Download dashboardbase-mcp.mcpb from the latest release and drag it into Claude Desktop's extensions settings. The bundle is self-contained โ you don't even need Node.
Cursor, Windsurf, and most other clients
These use the mcpServers shape. Add it to ~/.cursor/mcp.json (Cursor), ~/.codeium/windsurf/mcp_config.json (Windsurf), or your client's equivalent:
VS Code is the exception โ it uses servers rather than mcpServers, and requires an explicit type. In .vscode/mcp.json (workspace) or your user profile:
Requires Node.js 20 or newer โ that's the only prerequisite, and only for the npx routes. Nothing else to configure: the validation API is public, so there's no account, key or environment variable to set.
Staying up to date
Two different things arrive by two different routes, which is worth knowing:
Better error messages arrive on their own. The server holds no schemas โ it calls the live API โ so as validation messages improve, you see them immediately. Nothing to update.
New tools need a package update. If a release adds a tool, you get it once the package updates and your client restarts the server. A running server's tool list is fixed for the session.
How you installed
How you update
npx (Claude Code, Cursor, VS Code, Windsurf, Codex)
Automatic โ restart your client and the next server start picks up the newest version.
Claude Desktop .mcpb
Pinned at install. Download the new bundle from releases and drag it in again.
Docker
docker pull / rebuild the image.
Check what you're running:
bash
npx -y @dashboardbase/mcp --version
If that reports an older version than the latest release, npm is serving cached registry metadata โ a just-published version can take a few minutes to show up. Force it:
bash
npx -y @dashboardbase/mcp@latest --version
Releases follow semver: new tools are a minor bump and never break existing calls, so updating is safe by default.
Using it
First, check it's connected
bash
claude mcp list
You should see dashboardbase listed as connected. Inside a Claude Code session, /mcp shows the server and its two tools. In other clients, look for dashboardbase in the MCP or tools panel.
If it isn't there, the server never started โ check that node --version is 20 or newer, and that your config file has valid JSON.
Then just ask
You don't call these tools yourself. Describe what you want and the agent decides to call them, the same way it decides to read a file. There's no special syntax and nothing to remember.
Validating something you already have:
"Validate my dashboardbase setup file"
"Is .dashboardbase/revenue.json valid?"
"Check this against the dashboardbase contract" โ then paste the JSON
Validating as part of building:
"Build a KPI endpoint for MRR, then validate the response it returns"
"Add a gauge widget to my setup file and make sure it still validates"
Debugging something broken:
"My gauge widget shows an error in dashboardbase โ check what my endpoint returns against the contract"
"Why is dashboardbase rejecting my setup file?"
What it looks like
You: My revenue widget isn't rendering. Check .dashboardbase/revenue.json.
Agent:calls validate_setup_file
code
Invalid setup file โ 2 errors
mappings[0].type 14:9 Unknown widget type "guage"
mappings[1].path 22:5 Path must start with "/"
Agent: Two problems โ guage is a typo for gauge, and the second mapping's
path is missing its leading slash. Fixing both now.
edits the file, calls validate_setup_file again
code
Valid setup file โ no errors.
That loop โ validate, fix, re-validate โ is the whole point. The agent closes it without you opening dashboardbase.
Or invoke it directly
In Claude Code, the server adds a slash command โ type / and it's in the menu:
Leave the path off and it validates whatever you're working on, or whatever you paste next. This is the deterministic route: it doesn't depend on the agent deciding the tool is relevant.
The middle segment comes from the name you installed under. If you ran claude mcp add dashboardbase โฆ it's /mcp__dashboardbase__validate; install it as something else and that segment changes to match.
In clients that don't support MCP prompts, just name the tool in plain language:
"Use validate_setup_file on .dashboardbase/revenue.json"
"Use validate_widget_response to check this"
Agents sometimes need the nudge the first time; after that they usually pick it up on their own.
The two tools
Tool
Arguments
Notes
validate_setup_file
content โ the file text path โ read from disk instead
Give one or the other.
validate_widget_response
response โ the endpoint's JSON body path โ read from disk instead widgetType โ optional, e.g. kpi
Inferred from the response when widgetType is omitted.
Two things worth knowing:
path only exists over stdio. A hosted HTTP deployment has no access to your disk, so the parameter isn't offered there โ pass content / response instead.
validate_widget_response wants the full response body โ the title / actions / data / alert envelope your endpoint actually returns, not just the inner data payload.
Configuration
Everything is optional.
Variable
Default
Purpose
DASHBOARDBASE_API_URL
https://api.dashboardbase.com
Point at a different environment.
DASHBOARDBASE_API_KEY
unset
Sent as x-api-key. Not needed for the public API.
DASHBOARDBASE_TIMEOUT_MS
15000
Request timeout in milliseconds.
Running it as an HTTP server
For containers or a shared internal deployment:
bash
npx @dashboardbase/mcp --http --port 3000
# or
docker build -t dashboardbase-mcp . && docker run -p 3000:3000 dashboardbase-mcp
Serves Streamable HTTP at /mcp and a health check at /health. It's fully stateless, so it scales horizontally with no session affinity. Browser origins are refused unless you allowlist them with --allowed-origin https://example.com. The file-reading path parameter is not exposed in this mode.
What's in the repo
code
mcp/
โโโ src/
โ โโโ index.ts # CLI entry โ stdio by default, --http optional
โ โโโ server.ts # Registers the two tools
โ โโโ api.ts # Client for the dashboardbase Tools API
โ โโโ format.ts # Renders results as readable text
โ โโโ http.ts # Stateless Streamable HTTP handler
โ โโโ tools/ # One file per tool
โโโ test/ # Unit tests, no network required
โโโ tools.json # OpenAPI spec for the validation API
โโโ server.json # MCP Registry manifest
โโโ manifest.json # Claude Desktop bundle manifest
โโโ Dockerfile
How validation works
The server doesn't carry a copy of the schemas โ it calls the live dashboardbase validation API. So it can't drift from what the platform accepts, and error messages get better as the API does, with nothing to upgrade on your side.
That applies to the validation rules and messages, not to the server itself โ a release that adds a tool still needs a package update. See Staying up to date.
Related
dashboardbase โ the product. Build, host, and share dashboards from your APIs.
dashboardbase skill โ teaches your agent the JSON contract so it writes correct endpoints in the first place. Pairs directly with this MCP.
Found a gap or a confusing error? Open an issue. Note that the validation rules themselves live in the dashboardbase backend โ if a message is unclear that's still worth reporting here, and we'll fix it upstream.
The MIT license covers the code in this repo. "dashboardbase" is a trademark of dashboardbase โ see dashboardbase.com. You're free to use and adapt the server; please don't use the name or branding in a way that implies official affiliation.
Install
Configuration
Environment variables
DASHBOARDBASE_API_URL
Base URL of the Dashboardbase API. Defaults to https://api.dashboardbase.com.