Charts Your Agent Can Actually Read¶
Most charting libraries are built for human eyes. BioMCP's are built for both.
When an AI agent generates a PNG chart, it can't read its own output. It rendered pixels it has no way to parse back into numbers. The agent needs a vision model, a screenshot, and a prayer.
BioMCP takes a different approach. Every study command that produces data can also produce a chart — in three formats, each designed for a different consumer:
- Terminal renders the chart inline using Unicode block and Braille characters. The agent sees it in its own context window. No file, no screenshot, no round-trip.
- SVG is structured XML. An agent can parse exact numeric values directly from element attributes — 100% accuracy, no vision model needed, 36x smaller than PNG.
- PNG is for humans sharing charts in presentations and social media.
The charting engine is Kuva, an open-source Rust library with 29 plot types. BioMCP links it directly — no Python, no R, no subprocess. Charts compile into the single binary.
Why this matters for agents¶
A coding agent running BioMCP in a terminal session can query data and see the chart in the same output stream. It doesn't need to open a file, switch windows, or invoke a vision model.
That's a Kaplan-Meier survival curve drawn with Braille characters directly in the terminal. The agent can see the separation between TP53-mutant and wildtype groups, reason about the pattern, and immediately run the next command.
When the agent needs precision, SVG is the better output:
From height="405.9" and the axis scale, the agent recovers the exact count: 3,157 missense mutations. No estimation. No OCR. The TP53 mutation bar chart is 4.9 KB as SVG versus ~175 KB as PNG — 36x smaller with higher fidelity.
Eight chart types¶
Add --chart <type> to any study command. BioMCP validates the combination — ask for a violin plot from a mutation query and it tells you what's valid instead of producing garbage.
Survival curves¶
TP53-mutant patients: median 21.0 months. Wildtype: 32.1 months. p = 9.40e-29.
Expression distributions¶
biomcp study query --study brca_tcga_pan_can_atlas_2018 --gene ERBB2 \
--type expression --chart histogram -o erbb2-histogram.svg
The bimodal distribution is the signature of HER2-positive breast cancer. The right-hand bump is the ~15-20% of breast cancers with HER2 amplification.
Violin and ridgeline plots¶
biomcp study compare --study brca_tcga_pan_can_atlas_2018 \
--gene TP53 --type expression --target ERBB2 \
--chart violin -o erbb2-by-tp53-violin.svg
ERBB2 expression stratified by TP53 mutation status. The violin reveals the full distribution shape — the bimodal HER2 pattern is visible in both groups.
Ridgelines stack the same density curves vertically for easier visual comparison.
Mutation and co-occurrence charts¶
biomcp study query --study msk_impact_2017 --gene TP53 --type mutations \
--chart bar -o tp53-mutation-bar.svg
3,157 missense mutations dominate TP53 in MSK-IMPACT, followed by 683 nonsense and 517 frameshift deletions. Same data, two renderings — SVG for sharing, terminal for exploration.
Themes and accessibility¶
Four themes and twelve color palettes, including five designed for colorblind accessibility:
biomcp study query --study msk_impact_2017 --gene TP53 --type mutations \
--chart bar --theme dark --palette wong
| Themes | Accessible palettes |
|---|---|
light, dark, solarized, minimal |
wong, okabe-ito, deuteranopia, protanopia, tritanopia |
What's next¶
BioMCP currently uses 8 of Kuva's 29 chart types. Heatmaps for co-occurrence matrices, scatter plots for two-gene expression comparisons, and waterfall plots for mutation burden are on the roadmap.
Try it¶
uv tool install biomcp-cli
biomcp study download msk_impact_2017
biomcp study survival --study msk_impact_2017 --gene TP53 \
--chart survival --terminal
Full chart reference: Chart Documentation.