Rotifer Protocol

Posted on Mar 25 • Edited on Apr 6 • Originally published at rotifer.dev

Introducing @rotifer/mcp-server

#ai #opensource #mcp #devtools

The Rotifer gene ecosystem is now accessible to any AI agent that speaks MCP. One command, zero configuration:

npx @rotifer/mcp-server

Add it to your MCP client config and your agent can search, inspect, execute, and compare genes from the cloud registry — directly within conversations.

Why MCP?

The Model Context Protocol (MCP) is becoming the standard interface between AI assistants and external tools. Rather than building custom integrations for each AI platform (Claude, Cursor, Windsurf, ChatGPT), MCP provides a single protocol that all platforms converge on. By implementing an MCP server, Rotifer's gene ecosystem becomes instantly available in every MCP-compatible environment.

The key insight is that genes are inherently tool-shaped. Each gene has a defined input schema, a defined output schema, and a pure function that transforms one into the other. This maps perfectly to MCP's tool abstraction. What the CLI does via rotifer run, the MCP server does via tool calls — same genes, same execution, different interface.

What It Does

@rotifer/mcp-server exposes Rotifer's Cloud API as MCP tools and resources. Your AI agent gets:

7 Tools

Tool	Description
`list_genes`	Search and browse the gene registry with filters
`get_gene`	Get detailed gene info including phenotype and README
`run_gene`	Execute a gene with custom input
`compare_genes`	Side-by-side comparison of two genes
`get_gene_stats`	Download statistics by time period (7d/30d/90d)
`get_leaderboard`	Developer reputation rankings
`get_developer_profile`	Developer profile and reputation data

Tool Design Rationale

The seven tools were chosen to cover three agent workflows:

Discovery (list_genes, get_gene, get_leaderboard) — an agent exploring what capabilities exist in the ecosystem.
Evaluation (run_gene, compare_genes, get_gene_stats) — an agent deciding which gene best solves a specific problem.
Attribution (get_developer_profile) — an agent attributing results to their creators.

Each tool returns structured JSON with consistent error formats, so agents can programmatically handle failures without parsing error messages.

5 MCP Resources

Resource URI	Description
`rotifer://genes/{id}`	Gene details
`rotifer://genes/{id}/stats`	Gene statistics
`rotifer://developers/{username}`	Developer profile
`rotifer://leaderboard`	Global leaderboard
`rotifer://local/genes`	Scan local workspace for installed genes

Local Gene Discovery

The list_local_genes tool scans your workspace for installed genes and returns metadata, compile status, and cloud origin — bridging your local development with the cloud registry.

Real-World Usage Scenarios

Here are three scenarios where the MCP server changes the development experience:

Scenario 1: Gene Selection in Conversation
You're building an agent and need a text summarization gene. Instead of leaving your editor to browse the registry, you ask your AI assistant: "Find me the best summarization gene." The assistant calls list_genes with domain filter text.summarize, compares the top candidates with compare_genes, and recommends one based on fitness scores and download counts.

Scenario 2: Inline Gene Testing
You've written a new gene and want to verify it works. Your assistant calls run_gene with test inputs and shows you the output — all within the conversation. No terminal switching, no manual rotifer run commands.

Scenario 3: Ecosystem Health Monitoring
A project maintainer asks: "How are my published genes performing?" The assistant calls get_developer_profile and get_gene_stats for each gene, producing a summary of download trends, reputation changes, and Arena rankings.

Setup

Cursor

{
  "mcpServers": {
    "rotifer": {
      "command": "npx",
      "args": ["@rotifer/mcp-server"]
    }
  }
}

Claude Desktop

{
  "mcpServers": {
    "rotifer": {
      "command": "npx",
      "args": ["@rotifer/mcp-server"]
    }
  }
}

Works with any MCP-compatible client. Zero API keys required — the server uses the public Cloud API by default.

Custom Endpoints

Self-hosting or using a regional mirror? Set your endpoint:

ROTIFER_CLOUD_ENDPOINT=https://your-instance.example.com npx @rotifer/mcp-server

Or configure via ~/.rotifer/cloud.json.

The Journey: v0.1 → v0.2 in 4 Days

Version	Date	Milestone
v0.1.0	Mar 11	Initial release: 4 tools, Cloud API integration
v0.1.1	Mar 12	Fix npx binary resolution, query sanitization
v0.1.2	Mar 13	README with setup guides for major MCP clients
v0.2.0	Mar 14	3 new tools, 5 MCP resources, 139 tests

By the Numbers

7 MCP tools
5 MCP resources
139 tests (unit, integration, protocol, security, resilience)
Zero-config setup with public Cloud API defaults

Get Started

npx @rotifer/mcp-server

Looking Ahead

The MCP server bridges two worlds: the Rotifer gene ecosystem and the rapidly growing landscape of AI assistants. As more AI platforms adopt MCP, the server becomes a universal adapter — genes published once are accessible everywhere. Future versions will add mutation tools (publish, rate, flag) alongside the current read-only operations, enabling AI agents to not just consume genes but contribute to the ecosystem.

DEV Community