
mcp
com.outputcharts/mcp
Dashboards from JSON with a stable public link; sign up, build and keep fresh, all as tools.
Documentation
Output Charts skill for AI agents
Output Charts turns JSON into a dashboard with a stable public link. This repository holds the skill file that teaches an agent the whole flow: open an account for the person it works for, build a published dashboard in one HTTP call, hand back the link, and keep the page fresh with pushes from a script, a cron job or its own next run.
SKILL.md is a copy of the file the site serves at outputcharts.com/skill/SKILL.md. The site's copy is the one that moves first; ./check-skill.sh says whether this one still matches it.
What the skill does
Given a person who wants "a dashboard of these numbers", an agent that has read SKILL.md can:
- Sign the person up with
POST /v1/signup(no key needed): one call with their email address answers a Manage key, an MCP URL and a claim link. The account starts on a seven-day Solo trial, no card. - Build a live dashboard in one call:
POST /v1/dashboardswith the visuals andpublish: trueanswers the public URL. Re-running the same body never makes duplicates. - Keep it fresh: replace one visual, append rows to a chart or a table (JSON or CSV), or update several visuals at once by ref.
- Make a narrower key for CI or a cron, so the Manage key never lands in a repository secret.
- Handle the end of the trial: a failed call answers
requires_solowith afix; the agent asks for a Stripe Checkout link and hands it to the person.
Four worked examples are valid as written: a pull request queue, AI coding spend, a client's page speed report and a coding team's flow metrics.
Install it
Claude Code
Claude Code loads skills from .claude/skills/<name>/SKILL.md, in a project or in your home folder. Either way, one command:
mkdir -p ~/.claude/skills/outputcharts && curl -fsSL https://outputcharts.com/skill/SKILL.md -o ~/.claude/skills/outputcharts/SKILL.md
or, for one project, the same into .claude/skills/outputcharts/SKILL.md inside the repository. From then on, "make me a dashboard of …" uses it, and /outputcharts invokes it by name.
Any other agent
The file is plain Markdown with a two-line front matter, so it works wherever an agent reads instructions:
- Put it in the repository the agent works in (for example
docs/outputcharts-skill.md) and point to it fromAGENTS.md,.cursor/rulesor the agent's system prompt. - Or paste its contents into the agent's instructions.
- Or give the agent the URL
https://outputcharts.com/skill/SKILL.md; agents that read llms.txt find it there too.
The MCP server
The same API is an MCP server at https://api.outputcharts.com/mcp (streamable HTTP). It works before an account exists: with no key it serves two tools, signup and describe_visual_types. After signup, reconnect with the key as a bearer token, or add the per-key URL the sign-up returns (https://api.outputcharts.com/mcp/<key>) as a connector in a chat app that cannot send a header.
Claude Code:
claude mcp add --transport http output-charts https://api.outputcharts.com/mcp --header "Authorization: Bearer $OC_KEY"
Cursor, Claude, ChatGPT and Grok steps are in the MCP server docs. The full server description for directories is in mcp-server.md, and server.json is the entry for the official MCP Registry.
Links for agents and people
- llms.txt: the one file an agent needs to find everything.
- Agent API docs: every call, every error with its fix.
- OpenAPI 3.1 and the visual types: no key needed.
- Pricing: Free is one dashboard by paste; Solo is $10 a month or $96 a year for 20 dashboards, API keys, push and Format with AI. Every new account starts with seven days of Solo, no card.
Keep this copy current
./check-skill.sh
fetches the site's file and diffs it against SKILL.md; exit code 0 means they match. Pass a path to compare with a local copy instead.
License
MIT, see LICENSE. The Output Charts name and mark belong to Output Charts.
No installation packages available.