Harness Council
Local multi-harness orchestration for Claude Code, Grok CLI, and AGY CLI, enabling spawning, management, and coordination of AI coding agents in a visible terminal grid.
README
Harness Council
Local multi-harness orchestration for Claude Code, Grok CLI, and AGY CLI — spawn AI coding agents from different vendors as visible terminal workers, pass context between them, and watch them work in a 2×2 grid.
Council owns jobs + shared context. Harnesses own intelligence. cmux owns visibility.
You + Claude (plan) ──spawn──► AGY (explore) / Grok (code)
▲ │
└──────── result + notify ─────┘
(visible splits in cmux)
Guides
| Doc | Contents |
|---|---|
| docs/SETUP.md | Install, daemon, cmux, Claude MCP, plugin, config, troubleshooting |
| docs/USAGE.md | Roles, CLI/MCP, explore→code flow, 2×2 layout, auto-close |
| docs/ORCHESTRATOR.md | Any harness as the top orchestrator — AGENTS.md/GROK.md snippet, routing precedence |
Requirements
| Tool | Role | Notes |
|---|---|---|
| Node.js 22+ | runtime | uses node:sqlite |
| cmux | visibility | recommended — splits/tabs for workers |
| Claude Code | orchestrator / worker | optional |
| Grok CLI | worker | optional, grok login |
| AGY CLI | worker | optional |
Only the harnesses you actually route to need to be installed — any one of them can be the top orchestrator (docs/ORCHESTRATOR.md).
Quick start
git clone https://github.com/dorukanc/harness-council.git
cd harness-council
npm install && npm run build
npm link # or: export PATH="$PWD/bin:$PATH"
council setup
council daemon start
council spawn --role explore "Map the auth system"
council spawn --role code --effort high "Implement rate limiting"
council list
council result <job_id>
What you see in cmux
Default layout is a 2×2 grid of splits, then tabs if more workers start:
- First worker → split right
- Next → split down
- Up to 4 worker panes (2×2)
- More → tabs on worker panes
When a job completes successfully, its surface auto-closes so space frees for the next worker. Failed jobs stay open for inspection.
[display]
layout = "grid"
grid_max = 4
close_on_complete = true
close_on_failed = false
Claude Code
MCP (recommended):
{
"mcpServers": {
"harness-council": {
"command": "council",
"args": ["mcp"]
}
}
}
Optional plugin: plugins/claude-code (skill + /spawn + stop-hook). Details in docs/SETUP.md.
Register the MCP server once at user scope and every project works with no per-project setup — no CLAUDE.md line required. The council-runtime skill loads itself when you ask for delegation.
Per-project overrides (optional)
Only when a repo should route differently from your defaults:
council init # writes .harness-council.toml + a CLAUDE.md block
council init --file AGENTS.md --no-toml
Both writes are idempotent — re-running updates the managed block between
<!-- harness-council:start --> / <!-- harness-council:end --> and leaves the rest of
your file alone. The generated TOML is fully commented: it documents what the project
resolves to today, and stays inherited from ~/.harness-council/config.toml until you
uncomment a key.
Default roles
| Role | Harness | Purpose |
|---|---|---|
plan |
claude | Planning |
code |
grok | Implementation |
explore |
agy | Cheap scout / research |
review |
grok (read-only) | Second pass |
Configure in ~/.harness-council/config.toml or project .harness-council.toml.
Architecture
council CLI / MCP → local daemon → SQLite jobs + context store
→ adapters (claude | grok | agy)
→ cmux grid layout + auto-close
Development
npm test
npm run typecheck
npm run build
npm run council -- setup
How the daemon talks to cmux
cmux's default socket auth (cmuxOnly) verifies process ancestry, so a detached daemon can never open splits. When started from inside cmux, the daemon therefore runs in its own council · daemon tab; closing that tab stops it. Outside cmux everything still works headlessly.
Status
v0.1 — daemon (cmux-tab hosted), CLI, adapters, grid layout, auto-close, MCP, Claude plugin (local marketplace install), orchestrator-agnostic docs.
Contributing
Issues and PRs welcome. Before submitting:
npm run typecheck && npm test
Adding a new harness is intentionally small: implement the HarnessAdapter interface in src/adapters/ (binary resolution, readiness, command building, result parsing) and register it in src/adapters/index.ts.
License
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.