Developer Documentation

API, MCP, scoring, and data trust for agent project selection.

Git.Top is an agent-native knowledge layer for GitHub repositories. It turns project metadata, repository signals, deployment hints, alternatives, and quality evidence into structured responses that agents can cite and compare.

Connect an agent Try REST search Open workflow Agent quickstart Agent recipes Integration paths Inspect MCP tools Public benchmark Check quality governance

Agent Quickstart

Shortest path to a useful answer

curl "https://git.top/api/search?q=cloudflare%20agent&limit=5"
curl "https://git.top/api/project/cloudflare/agents"
curl "https://git.top/api/compare?repos=cloudflare/agents,langchain-ai/langchain&deployment=cloudflare"

Production agents should inspect metadata.source, classification, and quality_signal_confidence before presenting a recommendation as high-confidence.

Agent Surface Map

Human pages, REST, and MCP stay aligned

  • /api/agent-map maps each Git.Top concept to its page, REST endpoints, MCP tools, output fields, and trust fields.
  • Read short_path first, then reference_path only when you need the fuller discovery surface.
  • Use it when deciding whether to call project lookup, recommendations, alternatives, graph, compare, score, Atlas, GRP, or quality endpoints.
  • The same map is exposed through GET /mcp discovery for agent runtimes.

Agent Workflow

Turn intent into a selection path

  • /workflow and /api/workflow expose a guided path from trends to shortlist, graph, alternatives, score, compare, and trust policy.
  • get_agent_workflow, get_atlas, and get_quality_report make workflow, Atlas, and corpus quality available through MCP.
  • Start here when the task is broad, ambiguous, or likely to benefit from trend context before a direct recommendation.

MCP

Tools agents can call

  • search_projects for retrieval by query, category, deployment, language, and Cloudflare readiness.
  • get_project for structured project knowledge and evidence.
  • get_agent_workflow, get_atlas, and get_quality_report for guided selection, ecosystem maps, and corpus trust.
  • compare_projects and git_top_grp_query for graph-grounded selection.

Scoring Methodology

Quality and agent score are separate

Quality score

40% 30-day stars, 20% commits, 15% releases, 15% contributors, 10% issue response.

Agent score

22% documentation, 24% maintenance, 20% deployment, 18% popularity, 16% community.

Data Freshness

Trust the metadata, then the result

  • metadata.source: d1 means live indexed data; seed means fallback.
  • Project records include project.synced_at and metric records include metrics.calculated_at.
  • /status renders D1 availability, sync freshness, cursor progress, and recent sync runs for operators and agents.
  • /api/sync/status reports freshness, sync health, cursor progress, and recent failures.

Evidence Model

Recommendations should be explainable

Classification evidence separates category, deployment, difficulty, and Cloudflare readiness. Quality signal confidence marks estimates, partial counts, unknown counts, and snapshot-backed star movement.

Quality Governance

Review queues become auditable corrections

  • /quality/review shows low-confidence classifications, collection semantics, and signal gaps.
  • /coverage shows corpus boundaries, category distribution, and collection representation before agents rely on broad recommendations.
  • /api/quality/review exposes the same backlog for agent and operator workflows.
  • /api/admin/classification-overrides is protected by SYNC_SECRET and records reviewed one-off corrections separately from generated Agent Cards.

Site Trust

Production surfaces

  • Canonical host: https://git.top
  • Security contact: security@git.top
  • Machine discovery: /robots.txt, /sitemap.xml, /status, /trust, /benchmark, /quality, /coverage, /llms.txt, /llms-full.txt, /.well-known/security.txt, /mcp
  • Integration contact path: /integrations summarizes REST, MCP, GRP, production checks, boundaries, and security contact.