A Field Guide to 154 Agent Harnesses, Readable by Humans and Agents

What it does The repository maintains a curated, ranked list of 154 AI agent harnesses and orchestration frameworks, updated weekly. It maps the ecosystem with searchable tables, use-case indexes, and head-to-head comparison guides. Everything is backed by machine-readable exports—JSON, llms.txt, and an MCP server—so coding agents can query the list and recommend harnesses to users.

The interesting bit The project treats its own subject matter recursively: it builds an MCP server so agents can consume the catalog, effectively making the list self-referential. The rankings weight factors like autonomy regime, failure recovery, and sandboxing rather than treating star count as gospel.

Key highlights

• 154 projects tracked across categories from coding agents to multi-agent orchestration and memory layers
• Interactive visualizations plotting adoption surface area vs. stars, and autonomy vs. recovery tiers
• Use-case-driven index (“I want a turnkey coding agent today”) instead of pure taxonomy browsing
• MCP server with recommend, compare, and pick_harness tools, published to the official MCP registry
• Weekly rescoring with graveyard warnings and flags for star manipulation

Caveats

• Rankings mix subjective “harness concern” weighting with activity metrics, so the ordering is editorial, not empirical
• The “post–June 2026 billing reality” mentioned in decision guides is unexplained in the provided sources
• The list is broad enough that some entries blur the line between harness and general framework

Verdict Worth bookmarking if you are trying to ship an agent and need to narrow 154 options to a shortlist; skip it if you already know your stack and just want implementation docs.