Ennodia
MCP server for multi-agent review with Compare and traceable receipts
No single model or agent should be the only reviewer for work that matters. Ennodia lets your primary agent ask the installed agent CLIs you already have, track every child task, and use model-led Compare to surface agreements, disagreements, blind spots, and one synthesized answer with receipts.
The subscription-pool idea is secondary: if you already pay for Codex, Claude Code, Antigravity, OpenCode, and other local agent tools, Ennodia gives your main agent one MCP doorway to use them as an independent review panel. The experimental IO package exposes a small local HTTP subset for apps, but the supported first-class surface is MCP.
Install
Send this to your primary agent and let it handle setup:
try-ennodia.cherninlab.com
Or run it directly as a stdio MCP server from npm:
npx -y ennodia
Prefer a registry or client installer? Use the
Ennodia Smithery listing.
It uses the same stdio command from smithery.yaml (npx -y ennodia) and
does not need Ennodia-specific configuration.
Requires Bun 1.3.14 or newer — npx downloads Ennodia, Bun runs it. Prefer
Bun directly? Use bunx ennodia. For manual setup, local development,
or a full walkthrough, see
Quickstart.
What Ennodia does
- Discovers available local AI tools
- Plans a route with a caller-provided category or keyword fallback
- Estimates preflight input tokens and enforces local caps on that estimate
- Starts and monitors child tasks
- Shows status, timing, logs, and failures
- Cancels tasks and runs explicitly
- Compares multiple completed outputs
- Synthesizes one answer from the comparison
The main entrypoint is ennodia_run: it plans, executes, optionally
compares, and returns a run ID to poll with ennodia_get_run. See
MCP tools for the
full tool and parameter reference.
Ennodia is for deliberation-class work: a run usually takes minutes, and Compare adds two serial model passes after the child agents finish.
Ennodia IO
The separate @cherninlab/ennodia-io workspace package exposes a local HTTP and TypeScript interface for apps that want BYOK-style settings over installed local agents:
npx -y @cherninlab/ennodia-io
See Ennodia IO for supported fields, auth behavior, CORS posture, and current limits.
Supported harnesses
- Codex CLI
- Claude Code
- OpenCode
- Kilo Code
- Kiro CLI
- Cline CLI
- Hermes Agent
- Antigravity
Adapters stay thin — shared routing, tracing, task state, recovery, and Compare logic live in core modules.
Evaluated-but-not-shipped candidates include Gemini CLI, GitHub Copilot CLI, Amp, Aider, Goose, Qwen Code, and Cursor CLI. They will be added only when a supported non-interactive prompt-in/text-out surface can be verified without permission-bypass flags or provider-private APIs.
Documentation
- Installation for Agents — the agent-driven setup path
- Quickstart — manual setup and local development
- MCP Tools — full tool parameter reference
- How Ennodia Works — the orchestration pipeline
- Second Opinions — replicate, decompose, and red-team patterns
- Data Governance — local storage and data movement boundaries
- Comparisons — how Ennodia compares to adjacent tools
- Benchmarks — deterministic bug-recall results
- Running Better Audits — prompt rubrics for Compare
Benchmarks
The current benchmark is multi-model-bug-recall: small TypeScript review
fixtures scored against committed bug oracles. Run the deterministic suite with:
bun run bench:bug-recall
Live harness runs are available through bun run bench:bug-recall:live and are
kept out of bun run verify.
The current dated fixture snapshot is published in
Benchmarks: 4
cases, with ennodia-parallel-compare at 100% recall and 100% precision.
Contributing
Ennodia is under active development. Bug reports and small, focused pull requests are welcome. See CONTRIBUTING.md for the local verification workflow.
Server Config
{
"mcpServers": {
"ennodia": {
"command": "npx",
"args": [
"-y",
"ennodia"
]
}
}
}