Agent Harness Finder MCP Server: Search a Curated List of Agent Harnesses

• Gives one pick for your need

  Commits to one top harness for what you describe, with the reason, up to two alternatives, any to avoid and a guide to read next.

  Try: “What should I use for an always-on assistant in my chat apps?”

  Network · recommend

• Ranks harnesses for a use case

  Ranks harnesses from the curated list for a use case. You can cap how complex it is to adopt and require unattended runs or failure recovery.

  Try: “Pick a simple open-source harness for a terminal coding agent.”

  Network · pick_harness

• Picks tools across the agent stack

  Suggests picks for any layer, such as sandboxing, memory or evals. It also runs a live GitHub and Hacker News search and labels those results unvetted.

  Try: “Where can I run agent-written code safely?”

  Network · pick_infrastructure

• Compares 2 to 4 harnesses

  Lines up two to four projects on the list axes, says who leads where and warns if one is archived or flagged. It points to a matching guide.

  Try: “Compare openclaw/openclaw and NousResearch/hermes-agent.”

  Network · compare

• Finds and compares the best for a task

  Picks the top two to four harnesses for a task and returns them side by side with the reason each was chosen.

  Try: “Compare the best options for sandboxed code execution.”

  Network · compare_for

• Searches the list by keyword

  Searches names, descriptions, tags and categories across the list and returns matches sorted by relevance, then stars.

  Try: “Search the harness list for memory layers.”

  Network · search_harnesses

• Shows one project record

  Returns the full list record for one project by its GitHub id.

  Try: “Show the record for anomalyco/opencode.”

  Network · get_harness

• Lists the decision guides

  Lists the head-to-head guides with a slug, title and summary for each.

  Try: “What comparison guides exist?”

  Network · list_comparisons

• Reads one decision guide

  Returns the full text of one guide, fetched from the repo main branch.

  Try: “Read the OpenClaw vs Hermes guide.”

  Network · get_comparison

• Lists setup templates

  Lists copy-paste setup file sets, such as a shared AGENTS.md or safe Claude Code settings, with the files in each.

  Try: “What setup templates are available?”

  Network · list_templates

• Returns a template with its files

  Returns one template README plus the content and target path of every file, so your assistant can write them into your project.

  Try: “Get the safe Claude Code settings template.”

  Network · get_template

• Lists step-by-step guides

  Lists setup guides with a slug, title and summary, for example how to build your own harness.

  Try: “What playbooks do you have?”

  Network · list_playbooks

• Reads one playbook

  Returns the full text of one playbook.

  Try: “Show the playbook for building my own harness.”

  Network · get_playbook

• Lists installable skills

  Lists skills that check your own agent setup, with a description, install command and file list for each.

  Try: “Which skills can audit my agent setup?”

  Network · list_skills

• Returns a skill with its files

  Returns every file of one skill with the path to write it to, under .claude/skills/ or .agents/skills/.

  Try: “Get the tool-design-checker skill for Claude Code.”

  Network · get_skill

• Lists categories and scales

  Lists the categories and use-case intents with project counts, plus the complexity, autonomy and recovery scales.

  Try: “What categories does the list cover?”

  Network · list_categories

• Shows the weekly digest

  Shows what changed in the list: added and removed harnesses, biggest star gains, archived projects and new guides. Pass a date to catch up.

  Try: “What is new in agent harnesses since October 1?”

  Network · whats_new

Watch out for

• pick_infrastructure sends your text to GitHub and Hacker News

  The need text goes to api.github.com and hn.algolia.com. Set include_live_search to false, or avoid private details in the request. github.com

• Templates and skills return files for your assistant to write

  get_template and get_skill return file contents with target paths. The server writes nothing itself, but review the files before your assistant saves them into your project. github.com

• Live results are unvetted

  Results from pick_infrastructure live search have not passed the list curation bar. Treat them as leads. github.com

Security advisories