Flint
Microsoft Research's open-source chart language that compiles AI agent specs into Vega-Lite, ECharts, Chart.js, Plotly, or Excel.
How the 7.6 was reached
- Quality of output6/10
Paper shows Flint charts beating direct Vega-Lite generation and specs 85% shorter; a single research source, no broad adoption evidence. arxiv.org
- Ease of use8/10
One-line npm install and a getting-started guide taking users from spec to compiled chart. github.com
- Pricing value10/10
Free and open source under MIT with no paid tiers or required paid dependency. github.com
- API and integration quality8/10
Compiles to Vega-Lite, ECharts, Chart.js, Plotly and Excel formats; available as npm package and MCP server. github.com
- Solves the problem it claims to solve6/10
Users find it more reliable for standard chart types but less expressive than native Vega-Lite for bespoke charts. news.ycombinator.com
7.6/10 is the mean of the 5 criteria that apply. Scored on the five-criteria rubric v1, 29 September 2026. How scoring works
Pricing
- MIT license
- npm package (flint-chart)
- MCP server (flint-chart-mcp)
- Full language spec and reference implementation
Key Features
- Declarative ChartAssemblyInput schema AI agents generate without verbose parameterization
- Single spec compiles to Vega-Lite, ECharts, Chart.js, Plotly, and native Excel charts
- 70+ semantic field types (Rank, Temperature, Price, Country) capture data meaning rather than just shape
- Automatic layout adapts sizing, labels, marks, and legends to data cardinality
- MCP server lets AI agents create, validate, and render charts inside conversations
Pros & Cons
Pros
- Multi-backend compilation means one spec works across five chart targets including Excel
- MCP server wires into agent workflows without extra glue code
- 70+ semantic types reduce the guesswork LLMs face when describing axes and fields
- MIT license from Microsoft Research with active GitHub development
Cons
- Very new project with limited community examples and third-party integrations
- Host applications must implement rendering; the spec alone produces nothing
- ChartAssemblyInput schema has a learning curve compared to plain chart libraries
- Excel template count (18+) is narrower than the web chart backends
Flint addresses a real gap: LLMs generate verbose, brittle chart code when what they need is a structured intent format the host can render reliably. The five-backend compilation and MCP integration make it practical today, not theoretical. Whether it becomes standard depends on host adoption, and that work is still mostly ahead of it.
Try Flint →Added to scored.tools on
Competitors to Flint
Other tools in the coding category worth comparing.
Gradio
8.2/10Build and share machine learning web apps in Python with zero frontend experience required
Streamlit
8.4/10Open-source Python framework for building and sharing interactive data apps in minutes.
Marimo
8.6/10Next-generation reactive Python notebook that's Git-friendly and executable as scripts or apps