RyanAlberts/best-of-Agent-Harnesses

claude mcp add agent-harnesses -- uvx agent-harnesses-mcp  (needs uv)

Usage

• •Ask a coding/research agent to query the list via the MCP server: recommend (one pick + alternatives + what to avoid), compare/compare_for (2-4 harnesses side by side), pick_harness (ranked with complexity/autonomy/recovery filters), search_harnesses, get_harness, list_categories, list_comparisons/get_comparison.
• •Browse the searchable web site (ryanalberts.github.io/best-of-Agent-Harnesses/) with one page per harness, filterable by capability, autonomy, and recovery.
• •Point any agent at the raw llms.txt URL or harnesses.json for machine-readable data, or install the harness-scout / stack-auditor / harness-radar agent skeletons.
• •Use the comparison guides for decisions: how-to-pick-a-harness, OpenClaw vs Hermes, terminal coding agents, multi-agent orchestration, memory layers.

Key features

• ✓154 curated projects across 13 categories (progressive disclosure, coding agent products, harness configs/SDKs, personal agent runtimes, frameworks, multi-agent, plugins/MCPs/CLI tools, memory, eval/benchmarking, observability, research harnesses, libraries/SDKs).
• ✓Ranking guide with stars, simplicity/capability tiers, headless-ready and durable markers, license signals, capability tags, and concrete example links per project.
• ✓MCP server published to PyPI and the official MCP registry as io.github.RyanAlberts/agent-harnesses, with graveyard warnings for dead/star-manipulated repos.
• ✓Three open-source agent skeletons (harness-scout, stack-auditor, harness-radar) that run on existing AI subscriptions and deliver to Slack or Notion.

Best for

Developers and teams choosing an agent harness or orchestration framework, and agents that need an up-to-date, opinionated recommendation source.

Caveats

• ⚠This is a curated list/documentation project with an MCP query server, not a runnable agent harness itself.
• ⚠Star counts and rankings are editorial snapshots (captured 2026-08-02) and scores are from public docs.