| name | mcp-builder |
| description | Model Context Protocol server development for building platform extensions and tool integrations. |
| audience | platform |
| allowed_tools | ["file_read","file_write","bash_execute"] |
| metadata | {"category":"development","complexity":"high"} |
MCP Builder Skill
You build Model Context Protocol (MCP) servers to extend the MADFAM platform.
MCP Server Structure
An MCP server exposes tools, resources, and prompts to LLM agents:
my-mcp-server/
src/
index.ts # Server entry point
tools/ # Tool implementations
resources/ # Resource providers
prompts/ # Prompt templates
package.json
tsconfig.json
Tool Development
Each tool must define:
- name: Unique identifier (kebab-case).
- description: Clear explanation of what the tool does.
- inputSchema: JSON Schema for parameters.
- handler: Async function implementing the tool logic.
Best Practices
- Keep tools focused and single-purpose.
- Validate all inputs against the schema before processing.
- Return structured results (not just strings).
- Handle errors gracefully with descriptive messages.
- Log tool invocations for debugging.
- Rate-limit external API calls.
Integration with Selva
MCP servers in the MADFAM platform register via the nexus-api configuration.
Each agent can access MCP tools based on their skill permissions.
The permission engine evaluates tool calls through the standard HITL matrix.
