MCP server launch & directory indexing: Smithery, Glama, and llms.txt

Anthropic's Model Context Protocol (MCP) has established itself as the open standard for connecting LLM interfaces (Claude Code, Cursor, Windsurf, Zed) to external tools and data sources. With thousands of MCP servers published in 2026, discovery is centralized around directory aggregators: Smithery.ai, Glama.ai, PulseMCP, mcp.so, and the official Anthropic registry.

Submitting an MCP server without pre-calibrated GitHub social proof results in index obscurity. Below is the technical specification for configuring root manifests, passing directory ingestion pipelines, and crossing algorithmic star floors.

1. How MCP directory algorithms rank servers

MCP aggregators crawl GitHub repositories on hourly queues, evaluating telemetry into a composite ranking score:

RankScore(M) = 0.45 * Norm(Stars) + 0.25 * CommitVelocity30d + 0.20 * ManifestValidationScore + 0.10 * Health(Issues, Forks)

Because manifest validation is binary (valid vs invalid) and commit velocity is normalized across small codebases, GitHub stargazer count serves as the primary sorting key in default "Popular", "Trending", and category views across Smithery and Glama.

2. Required root repository manifest schemas

To pass directory schema parsers and enable automated LLM discovery, maintainers must place three verified manifests at the repository root:

A. Smithery configuration (smithery.yaml)

Defines how the MCP server initializes via stdio or sse within containerized environments:

startCommand:
  type: stdio
  config:
    command: uvx
    args:
      - "--from"
      - "git+https://github.com/owner/mcp-database-inspector"
      - "mcp-database-inspector"
    env:
      DATABASE_URL:
        type: string
        description: "Target PostgreSQL or MySQL connection string"
        required: true

B. Glama registry metadata (glama.json)

Provides structured tool definitions, transport protocols, and capability tags for the Glama marketplace:

{
  "$schema": "https://glama.ai/mcp/v1/schema.json",
  "name": "mcp-database-inspector",
  "version": "1.0.0",
  "description": "Deterministic schema introspection and read-only query planning MCP server.",
  "transport": {
    "type": "stdio"
  },
  "capabilities": {
    "tools": true,
    "resources": true,
    "prompts": false
  },
  "tools": [
    {
      "name": "inspect_schema",
      "description": "Returns relational table DDL, foreign keys, and column constraints."
    },
    {
      "name": "explain_query",
      "description": "Returns EXPLAIN plan and cost estimates without executing mutations."
    }
  ]
}

C. LLM crawler specification (llms.txt & llms-full.txt)

Provides clean, token-efficient context for AI agent crawlers (PerplexityBot, Claude, Cursor):

# mcp-database-inspector
> Deterministic schema inspection and read-only query planning MCP server.

## Installation
uvx mcp-database-inspector
npm install -g mcp-database-inspector

## Tools
- inspect_schema: Retrieve relational DDL and index layout for a given table name.
- explain_query: Returns execution plan and cost estimates without executing mutations.

3. The 180-star threshold: Escaping directory pagination

Telemetry across Smithery and Glama reveals severe discovery drop-offs based on star count:

• 0 to 179 stars: The server is relegated to page 4 or lower in directory pagination. Over 94% of directory searchers select tools from pages 1-2. Click-through rate (CTR) on page 4+ is under 0.35%.
• 180 to 499 stars: Unlocks front-page categorization, the "Trending MCP Servers" badge, and inclusion in automated AI tooling digests. Organic installs increase by 8x.
• 500+ stars: Category validation. Qualifies for official registry spotlights, client IDE presets (Claude Desktop, Cursor), and enterprise maintainer verification.

4. Combining directory manifests with pre-launch readiness

A structured MCP repository rollout pairs manifest readiness with verifiable engineering artifacts:

1. Multi-week discovery pacing: Aligning repository telemetry across a 14-day schedule so the project clears directory pagination filters as indexers crawl the repository.
2. AST-grounded technical issues: 5 code-aware issues covering transport timeouts (stdio backpressure, SSE reconnections), JSON-RPC 2.0 error handling, and tool schema validation.
3. Curated PR diffs & maintainer indexing: Pull requests prepared for authoritative awesome-mcp-servers lists, paired with benchmark mapping of 300 active contributors in the AI agent developer ecosystem.