</>MCP Agents Market
Plugin

evo

by evo-hq1.4kPythonUpdated 2026-07-17

turns your codebase into an autoresearch loop — discovers what to measure, instruments the benchmark, then runs tree search with parallel subagents.

Claude CodeCodexCursorKimiOpenClawHermesOpencodePi

Evo is an autoresearch orchestrator plugin for agentic coding frameworks that automates code optimization through iterative experiments. It discovers benchmarks in your codebase, instruments evaluation metrics, and runs tree-search optimization using parallel sub-agents that propose changes, test them, and keep improvements. Compatible with Claude Code, Codex, Cursor, Kimi, OpenClaw, Hermes, Opencode, and Pi, evo coordinates multiple isolated workspaces where each sub-agent formulates hypotheses, edits code, and validates changes against configurable gates.

Key Features

Tree-search optimization over simple hill climbing, allowing multiple experiment branches to fork from any committed improvement
Parallel sub-agent execution with each running in isolated git worktrees, reading shared failure traces and discarded hypotheses
Automatic benchmark discovery that explores repositories, identifies what to measure, and instruments evaluation code
Configurable gates (regression tests, safety checks) that discard experiments failing to meet pass/fail criteria regardless of score
Multiple frontier selection strategies including argmax, top-k, epsilon-greedy, softmax, and Pareto-per-task for choosing which branch to extend
Cross-cutting RLM-inspired scans that surface compound failure patterns and shared root causes across experiment traces
Real-time dashboard for monitoring experiments, configuring backends, and adjusting frontier strategies
Flexible execution backends supporting local worktrees, SSH hosts, Modal, E2B, Daytona, AWS EC2, and Azure VMs

Use Cases

  • 01Optimizing performance of JSON parsers, compilers, or other compute-intensive components through automated iteration
  • 02Improving machine learning training loops by autonomously experimenting with hyperparameters and architecture changes
  • 03Refactoring legacy codebases while maintaining correctness through regression test gates
  • 04Exploring multiple optimization directions in parallel when the best approach is uncertain
  • 05Running benchmark-driven optimization on cloud infrastructure to leverage scalable compute resources
  • 06Discovering and instrumenting performance metrics in unfamiliar codebases before beginning optimization

Related Plugins

View more

andrej-karpathy-skills

A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls.

Plugin216.6k

claude-mem

Persistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More

Plugin95.2k

Understand-Anything

Graphs that teach > graphs that impress. Turn any code into an interactive knowledge graph you can explore, search, and ask questions about. Works with Claude Code, Codex, Cursor, Copilot, Gemini CLI, and more.

Plugin85k

rtk

CLI proxy that reduces LLM token consumption by 60-90% on common dev commands. Single Rust binary, zero dependencies

Plugin82.2k

evo — FAQ

What is the evo AI sub-agent?+

Evo is a plugin for agentic coding assistants that automates code optimization by discovering benchmarks, running parallel experiments in isolated workspaces, and keeping changes that improve metrics. It uses tree search with multiple sub-agents to explore different optimization paths simultaneously.

How do I install evo for Claude Code or Cursor?+

Install the evo CLI with 'uv tool install evo-hq-cli', install your host framework CLI (like '@anthropic-ai/claude-code' via npm), then run 'evo install <host>' where host is claude-code, codex, cursor, kimi, hermes, opencode, openclaw, or pi. The install command configures the plugin and necessary hooks automatically.

Which AI coding assistants work with evo?+

Evo supports Claude Code, Codex, Cursor, Kimi, OpenClaw, Hermes, Opencode, and Pi. Invocation syntax varies by host: '/evo:' prefix for Claude Code, '$evo' for Codex, skill menu on Cursor, and natural language on Hermes, Opencode, OpenClaw, and Pi.

Does evo require API keys or paid services?+

The evo plugin itself is free and Apache-2.0 licensed. You need a compatible agent host (some require API keys or subscriptions). Remote execution backends like Modal, E2B, Daytona, AWS, and Azure require accounts with those providers; local worktree execution has no additional requirements.

What are gates and why are they important?+

Gates are pass/fail checks (test suites, invariants, score floors) that run on every experiment to prevent evo from finding shortcuts like returning constants or trading correctness for speed. Experiments that fail gates are discarded even if their benchmark score improves.

Can I run evo experiments on cloud infrastructure?+

Yes, evo supports multiple remote backends including Modal, E2B, Daytona, AWS EC2, Azure VMs, and SSH hosts. Install with the matching provider extra like 'uv tool install evo-hq-cli[modal]' and configure the backend through the dashboard.

How do I install evo?+

Open the source repository on GitHub and follow its README. evo is a plugin — MCP Agents Market links you directly to the official repo.

Is evo free?+

evo is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.

Related searches

Developer Tools PluginsTesting PluginsProductivity & Workflow PluginsBest MCP ServersBest SkillsBest AgentsTop rated listingsClaude Code toolsBrowse all categories