Build your first PLASMA application
From zero to a living, evolving app in five minutes. No code to write, no IDE to open. Just a conversation with Claude.
- macOS
- Claude Desktop
- PLASMA Claude Components installed
Install PLASMA
PLASMA has three components to install: the PLASMA Connector MCPB (the bridge between Claude and your desktop), the PLASMA Surface OSX App (the rendering window where your apps come to life), and the PLASMA Plugin, a set of instructions that helps Claude use PLASMA capabilities in the most effective way.
Open Claude Desktop, go to Settings → MCP Servers, and add the PLASMA connector. The server handles all communication between Claude and the rendering surface.
Adding the PLASMA Connector MCPB to Claude Desktop. This enables Claude to create and control visual applications.
The PLASMA Plugin is a set of instructions that teaches Claude how to use PLASMA tools effectively. It enables optimized rendering, mutation strategies, and organism management. Install it from the Claude Desktop plugin interface.
The PLASMA Plugin gives Claude the knowledge to build, mutate, and evolve your applications.
The Surface is the desktop window where PLASMA renders your applications. Once active, it connects automatically to the PLASMA Connector.
The PLASMA Surface lives in the menu bar. Your applications render in a dedicated window, separate from the chat.
Start PLASMA
Open Claude Desktop and type:
/plasma:start
PLASMA checks the connection to the PLASMA Connector and the rendering surface, then lists all available tools. If both components show Connected, you're ready to go.
The /plasma:start command verifies that both the PLASMA Connector and the PLASMA Surface are connected and ready.
Use Cases & Ideas
open the excel file,
show data in a clean table,
browse it page by page,
click any cell to edit it,
search across all columns,
and download the updated file,
wait for events
The user drops an Excel spreadsheet into the conversation and asks Claude to build a visual explorer for it. Claude parses the file (columns, data types, row count) and generates a full interactive data viewer on the desktop: sortable table, column filters, search, and summary statistics.
No need to open Excel. The data is live on the PLASMA Surface, rendered as a responsive application the user can navigate, filter, and explore instantly.
The user asks for inline editing. Claude mutates the organism to add editable cells, validation, and a change tracker.
Claude reads the data shape, generates bar charts, distributions, and trend lines, and exports updated data back to Excel. All interactive, all from conversation.
PLASMA can generate virtually anything a browser can render, from data tools to creative experiments. Here are two examples of what emerges when you give Claude a surface and a prompt.
create a colorful 3D fractal environment on the plasma surface.
i want a mandelbulb rendered in real-time using raymarching with webgl.
the camera should orbit slowly around the structure.
colors should be vivid and cyclic (cosmic rainbow style),
and the fractal geometry should morph over time.
add three interactive sliders to control complexity (iterations),
animation speed, and color palette.
mouse movement should rotate the camera viewpoint.
dark background with stars and a luminous halo around the fractal.
clear colors, wait for events
i want a drag-and-drop workflow builder for AI agents.
think of it like a flowchart where each node is an agent
with its own prompt and model.
i should be able to add nodes, connect them by clicking
from one to another, delete them,
and see the full pipeline as JSON.
when i hit execute, send the workflow back to you to execute it.
wait for events
Ask Claude to build an environment for designing agentic configurations intuitively. It does it in under a minute. You draw your agentic architecture on the surface, Claude sees it, understands it, and executes it.
The user asks Claude to retrieve project engagements from Salesforce. Claude connects through a dedicated MCP connector, authenticates via SSO, and pulls live CRM records.
Instead of returning raw data in chat, Claude dynamically generates an interactive dashboard on the user's desktop: KPI cards, filterable tables, SLA indicators, progress bars, all shaped by the actual data retrieved.
The user filters, searches, drills into records. Claude watches every interaction. Clicking "Ask Claude" on an engagement triggers a contextual analysis: risks, next steps, dependencies.
As the conversation continues, the interface evolves. The user asks for a chart, Claude adds it. A new filter, a visual alert layer, a data export: everything appears in real time, driven by natural language.
Zero code, zero pre-built dashboards. Data source, visualization, and analytics, all orchestrated by Claude in a single conversational flow.
The user asks Claude to pull the current sprint from Jira Cloud. Claude connects through a dedicated MCP connector, authenticates via OAuth 2.0, and retrieves the full sprint payload: tickets, statuses, story points, assignees, blockers, and commit history.
Instead of listing raw issue data in chat, Claude dynamically generates a complete sprint board on the user's desktop: Kanban view, burndown chart, team workload distribution, and real-time counters for blockers, open PRs, and bugs.
The user clicks any ticket to inspect details. "Ask Claude" triggers contextual analysis of that issue. "Sprint Analysis" gives Claude full visibility on sprint health, surfacing uneven workloads, blocker chains, or burndown deviation.
As the conversation continues, Claude reshapes the board. A velocity trend, a priority heatmap, a filter by assignee, a CSV export: each request becomes an immediate UI update, no reload required.
A living sprint cockpit where Jira data, interactive visualization, and AI analysis converge in a single conversational flow. No dashboards to configure, no JQL to write, no plugins to install.
Download sample data used in this demo (VCF files, mock annotations).
A computational biologist uploads a VCF file from a whole-exome sequencing run and asks Claude to visualize the variants. Claude parses the file, cross-references against ClinVar and gnomAD databases through MCP connectors, and generates an interactive genome browser: a chromosome ideogram with variant density heatmap, a filterable variant table, and a lollipop plot showing mutation positions along key genes.
The researcher clicks a cluster of variants on chromosome 17. Claude zooms into the BRCA1 gene region, renders the exon-intron structure, and places each variant with its predicted functional impact color-coded by ACMG classification. Hovering a variant shows allele frequency, in-silico predictions (SIFT, PolyPhen), and links to published literature.
The researcher asks to compare against the control cohort. Claude overlays population frequencies, surfaces statistically significant variants, identifies compound heterozygotes in DNA repair genes, and renders a pathway diagram showing where each variant hits in the damage response network.
The complete analysis workspace: annotation, filtering, population comparison, and pathway mapping, all built interactively as the scientific question evolves.
Every PLASMA creation is automatically saved as an organism and can be reloaded instantly, without waiting for regeneration. All the changes you request through conversation (mutations) are saved incrementally and reapplied automatically when the organism is loaded again.
To see all your saved apps, just ask:
list my plasma apps
To reload a specific one:
load the plasma app excel_navigator
The organism loads in seconds with its full mutation history applied, exactly as you left it.
Depending on complexity, the initial generation can take anywhere from tens of seconds to a few minutes. The model that currently works best is Claude Opus, but many apps work beautifully with Sonnet, and simpler ones run well even with Haiku.
What you can build
PLASMA runs anything a browser can run. Some things people have built:
- Data entry forms with validation and database integration
- Real-time dashboards with Chart.js and D3.js
- 3D visualizations with Three.js and WebGL
- Generative art canvases with fragment shaders
- Interactive data explorers (SQL query → visual result)
- Landing pages and web applications
- Games and simulations
- Monitoring panels with live data feeds
Next steps
- Install PLASMA Claude Components
- Read the full article on how PLASMA works under the hood
- Hera Artificial Life, the multi-channel AI gateway with native PLASMA integration