AI Tools

LikeC4 provides two complementary tools for AI-powered development: Agent Skills for teaching AI assistants the DSL syntax, and an MCP Server for querying your architecture model.

Agent Skills

Agent skills are DSL references that AI coding assistants load automatically when editing .c4/.likec4 files. They teach agents correct LikeC4 syntax, patterns, and validation workflows — so they can write DSL code without hallucinating.

Install into any project

npx skills add https://likec4.dev/

This uses the Agent Skills Discovery protocol. Available skills:

| Skill | Description | |-------|-------------| | likec4-dsl | Complete DSL reference — specification, model, views, deployment, predicates, examples |

Works with Claude Code, Cursor, Windsurf, and other agents that support the protocol.

MCP Server

LikeC4 MCP Server provides knowledge of your LikeC4 model to LLMs. This enables you to query your model in natural language:

“Lookup LikeC4 model and list all incoming relationships of the backend api”

“What nested elements of the ‘Backend’ have relations with the legacy api”

“List all elements tagged legacy from team1 project”

“What technologies are used for ui (consider all elements with browser shape)”

“Export to CSV all relationships between Backend and Amazon SQS”

Usage

Three options are available:

Use extension’s built-in MCP Server
Use likec4 mcp CLI
Use @likec4/mcp package

Using extension

When LikeC4 extension is installed in VSCode, the MCP Server is registered automatically via VSCode’s native MCP support (stdio transport). No additional configuration is needed — the server appears in your MCP server list once the extension is active.

For other editors, start the MCP server via CLI and configure it manually:

Create .cursor/mcp.json:

{
  "mcpServers": {
    "likec4": {
      "url": "http://localhost:33335/mcp"
    }
  }
}

Then start the server: likec4 mcp --http

See Windsurf documentation for details:

{
  "mcpServers": {
    "likec4": {
      "serverUrl": "http://localhost:33335/mcp"
    }
  }
}

Then start the server: likec4 mcp --http

Add the MCP server to Claude Code:

claude mcp add likec4 -- npx -y @likec4/mcp

Or add to .mcp.json:

{
  "mcpServers": {
    "likec4": {
      "command": "npx",
      "args": ["-y", "@likec4/mcp"],
      "env": {
        "LIKEC4_WORKSPACE": "${workspaceFolder}"
      }
    }
  }
}

Using CLI

If you have installed likec4 CLI, you can start MCP server with stdio transport:

likec4 mcp
# or
likec4 mcp --stdio

Start MCP server with http transport on port 33335 (default) at ./src folder:

likec4 mcp --http ./src

Start MCP server with http transport on port 1234:

likec4 mcp -p 1234

Using `@likec4/mcp` package

Example configuration:

{
  "mcpServers": {
    "likec4": {
      "command": "npx",
      "args": [
        "-y",
        "@likec4/mcp"
      ],
      "env": {
        "LIKEC4_WORKSPACE": "${workspaceFolder}"
      }
    }
  }
}

This package starts MCP server using stdio transport.

If LIKEC4_WORKSPACE environment variable is not set, the current directory will be used as workspace.

You can also use @likec4/mcp package as CLI (this package is smaller than likec4):

npm install -g @likec4/mcp

# Start MCP server with streamable http transport
likec4-mcp --http --port 1234 /path/to/workspace

# See available options
likec4-mcp -h

Disable watch mode with --no-watch to consume less resources, if you have static workspace.

Available tools

| Tool | Description | |------|-------------| | list-projects | List all LikeC4 projects in the workspace | | read-project-summary | Project specification, configuration, all elements, deployment nodes and views | | search-element | Search elements and deployment nodes by id/title/kind/shape/tags/metadata | | read-element | Full element details including relationships, views, deployments, metadata | | read-deployment | Details of a deployment node or deployed instance | | read-view | Full view details (nodes/edges) and source location | | find-relationships | Direct and indirect relationships between two elements | | query-graph | Query element hierarchy (ancestors, descendants, siblings, children, parent) and single-hop relationships (incomers, outgoers) | | query-incomers-graph | Recursive BFS graph of all upstream dependencies/producers | | query-outgoers-graph | Recursive BFS graph of all downstream consumers/dependents | | query-by-metadata | Search elements by metadata key-value pairs (exact/contains/exists) | | query-by-tags | Tag filtering with boolean logic (allOf, anyOf, noneOf) | | query-by-tag-pattern | Search elements by tag patterns using prefix, contains, or suffix matching | | find-relationship-paths | Discover all multi-hop relationship chains between two elements via BFS | | batch-read-elements | Read full details for multiple elements in a single request | | subgraph-summary | Summarize descendants of an element with depth, metadata, and relationship counts | | element-diff | Compare two elements and show differences in properties, tags, metadata, and relationships | | open-view | Opens the LikeC4 view panel (editor only) |

Check out README for more details.