@coveord/plasma-mcp-server

MCP (Model Context Protocol) server for the Plasma design system. Gives AI agents dynamic access to component documentation — no need to load everything upfront.

For full coverage, pair this with @mantine/mcp-server: the Plasma server is authoritative for Plasma-wrapped components; the Mantine server covers the re-exported components and inherited props. Agents should query Plasma first and fall back to Mantine when a component isn't found.

Import invariant: always import from @coveord/plasma-mantine, even when Mantine docs were the reference. @coveord/plasma-mantine re-exports all Mantine components with Coveo's theme applied.

Tools

| Tool | Description | | --------------------- | ---------------------------------------------------------------------------------------- | | list_components | Returns all documented Plasma component names | | get_component_doc | Returns the full Markdown doc for a component (props, sub-components, design guidelines) | | get_component_props | Returns just the props table for a component | | search_docs | Full-text search across all component docs; returns the top matching excerpts |

Setup

Kiro

Option 1 — MCP servers (recommended for on-demand component docs):

Create or edit .kiro/settings/mcp.json in your project (or ~/.kiro/settings/mcp.json for all projects):

{
    "mcpServers": {
        "plasma": {
            "command": "npx",
            "args": ["-y", "@coveord/plasma-mcp-server"]
        },
        "mantine": {
            "command": "npx",
            "args": ["-y", "@mantine/mcp-server"]
        }
    }
}

Restart Kiro after saving the file. Both servers will appear in the MCP panel.

Option 2 — Steering file (always-on Plasma conventions, no MCP needed):

Create .kiro/steering/plasma.md and paste the contents of https://plasma.coveo.com/plasma-skill.md into it with this frontmatter:

---
inclusion: always
---

[paste plasma-skill.md contents here]

GitHub Copilot (VS Code)

Create or edit .vscode/mcp.json in your project:

{
    "servers": {
        "plasma": {
            "type": "stdio",
            "command": "npx",
            "args": ["-y", "@coveord/plasma-mcp-server"]
        },
        "mantine": {
            "type": "stdio",
            "command": "npx",
            "args": ["-y", "@mantine/mcp-server"]
        }
    }
}

Commit this file to share the MCP setup with your team. Use MCP: List Servers in the VS Code command palette to verify the servers are running.

Codex CLI

Add to ~/.codex/config.toml:

[mcp_servers.plasma]
command = "npx"
args = ["-y", "@coveord/plasma-mcp-server"]

[mcp_servers.mantine]
command = "npx"
args = ["-y", "@mantine/mcp-server"]

For a project-specific setup, set CODEX_HOME=.codex and create .codex/config.toml in your repo instead.

GitHub Copilot CLI (terminal)

The Copilot CLI uses skill files rather than MCP. Load the Plasma skill with:

/skill https://plasma.coveo.com/plasma-skill.md

Or add it to .github/copilot-instructions.md in your project to have it apply automatically.

Example prompts

Once the MCP server is connected, you can ask your AI agent:

"Show me all the available Plasma components."

"Get the documentation for the Button component."

"What props does the Modal component accept?"

"Search the Plasma docs for 'form validation'."

How it works

The component documentation is bundled at build time from @coveord/plasma-llms/dist/ into a dist/data.json file. The MCP server loads this file at startup and serves it via the MCP protocol — no runtime network calls or file I/O required.

For the underlying static files (useful for tools that fetch URLs directly), see @coveord/plasma-llms.