DeliberationBLOCK
Ask Codex, Gemini, Grok, and 400+ OpenRouter models (Qwen, Kimi, DeepSeek) for second opinions or arbiter-mediated consensus. One MCP server for Claude Code, Codex, Cursor, Kiro, OpenCode. Measures the models' usefulness, too.
Overview
From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.
Get a second opinion in Claude Code from GPT, Gemini, and Grok - plus 400+ more models through OpenRouter, including Qwen, Kimi, and DeepSeek. Seven domain experts (Architect, Code Reviewer, Security Analyst, and four more) review your plans, find bugs, and debate edge cases until they agree.
Recent blog post: Meet Deliberation: 400+ models is easy, knowing which ones earn a place is hard.
📸 See a full /consensus run: round 1 disagreement to round 5 convergence
... a few moments later ...
📸 See /ask-all stage a 2-round architect debate: three models, three verdicts, then each critiques the others - disagreement matrix included
When three models argue, the real bug reveals itself. Round 1 = independent top findings. Round 2 = each model dunks on the others' picks. The disagreement matrix shows where they diverge; the conclusion shows what to actually fix first.
📸 See the local dashboard draw a /consensus run as a timeline: one lane per voice, verdicts on each, final report alongside
The dashboard (/deliberation:dashboard, opt-in with "dashboard": { "enabled": true }) is a read-only page on 127.0.0.1. Live follows runs as they happen, *Runs
5f4b6d348806OBSERVED · 2026-10-07Connect
Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.
claude mcp add deliberation-mcp -- npx -y @antonbabenko/[email protected]
Exposed tools (18)
11 read · 7 write · 0 destructive.
| Tool | Risk | Description |
|---|---|---|
analyze | read | Analyze recent runs from the opt-in debug log (latency/tokens/reasoning-effort per model) plus the session store (verdict agreement rate), and return advisory tuning suggestions (disable a slow/redundant model in ask-all, lower an OpenRouter model |
ask-all | read | Fan out one question to GPT, Gemini, Grok, and any configured OpenRouter models in parallel for independent second opinions, then return all results (advisory, no cross-contamination). Pass |
ask-one | read | Second opinion from ONE named provider in the active panel (e.g. |
codex-login | write | Start (or join) the ChatGPT device login for GPT (Codex) and return its link + one-time code WITHOUT asking GPT anything. Use it when GPT has no login on this machine (e.g. Claude Code on the web): show the returned |
consensus | write | Run the FULL multi-round consensus convergence loop server-side with a provider arbiter (blind pass + peer fan-out -> adjudicate -> revise) and return the converged verdict. Default depth is |
deliberation | read | When and how to delegate to GPT, Gemini, Grok, and OpenRouter expert subagents via the deliberation MCP tools. |
gemini | write | Start a new Gemini expert session |
gemini-reply | read | Continue an existing Gemini session |
grok | write | Start a new Grok (xAI) expert session. Advisory only (no filesystem editing). Supports attaching files. |
grok-reply | read | Continue an existing Grok session (in-memory; lost if the MCP server restarts). |
openrouter | write | Start an OpenRouter expert session (advisory only). Pick a configured |
openrouter-list | read | List configured OpenRouter delegates and settings. Pass mode ( |
openrouter-reply | write | Continue an OpenRouter session by threadId (in-memory; lost on restart). |
panel | read | Return the names of the providers |
reload-mcp | read | Gracefully cycle dashboard and audit MCP processes after deliberation is updated. |
session-annotate | read | Append a freeform note to a persisted session |
session-get | read | Fetch a persisted consensus/ask-all session record by id (opinions, verdict, arbiter, annotations). Requires sessions.persist; local and read-only (no provider calls). Returns a text-wrapped JSON envelope { session }, or { error } when persistence is off or the id is unknown. |
session-revisit | write | Re-run a persisted session |
Trust audit
BLOCKgrade F · trust 41/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | PASS |
| L1 | Static analysis of the code | FAIL |
| L2 | Instruction surface (what it tells the agent) | FAIL |
| L3 | Class-specific surface | PASS |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- declared (18 observation(s))
- Network
- declared (1 observation(s))
- Shell
- declared (2 observation(s))
- Dependencies
- not all pinned
- Secrets in source
- found
Findings (25)
process.stderr.write("[deliberation] agy read-only run wrapped in sandbox-exec (workspace writes denied)\n");"**/.ssh/**",
"**/id_rsa", "**/id_rsa.pub", "**/id_ed25519", "**/id_ed25519.pub",
"**/id_ecdsa", "**/id_ecdsa.pub", "**/id_dsa", "**/id_dsa.pub",
"**/*.tfstate*", "**/.env", "**/.env.local", "**/.ssh/**", "**/*.pem", "**/*.key",
The script needs a POSIX shell (Git Bash on Windows). Without one, tell the user to run
3. **If none are available**: Do not delegate; inform the user that they need to run `/deliberation:setup`.
open the printed `http://127.0.0.1:<port>/?t=<token>` URL. It is a read-only browser view
open the printed `http://127.0.0.1:<port>/?t=<token>` URL. It is a read-only browser view
.replace(/^/, "")
.replace(/^/, "")
password: "superSecretPassword123",
api_key: "custom-proprietary-token-9988",
const ghHeader = "Authorization: Token ghp_ABCDEFGHIJKLMNOPQRSTUVWXYZ012345";
assert.equal(ghOut.includes("ghp_ABCDEFGHIJKLMNOPQRSTUVWXYZ012345"), false);const privKey = "-----BEGIN RSA PRIVATE KEY-----\nMIIEowIBAAKCAQEA0Yq123456789\n-----END RSA PRIVATE KEY-----";
const slackBot = "xoxb-123456789012-1234567890123-abcdef123456";
const slackEnt = "xoxe.xoxp-1-123456789012-abcdef123456";
test("RC4: an errored peer is excluded; remaining APPROVE -> converges", async () => {const { createJournal } = require("../../core/journal.js");const { makeRegistry } = require("../../core/registry.js");const { resolveRunsDir, resolveDashboardStatePath, resolveConfigPath } = require("../../core/paths.js");const { isSafeId } = require("../../core/journal.js");const { readSession, listSessions } = require("../../core/sessions.js");test("S3: advisoryEnv drops the kill-switch and scrubs push/exfil + credential-shaped env", () => {Gates applied: instruction_override, no_behavioural_pass.
5f4b6d348806full audit observations/trust-audit/mcp-server/antonbabenko__deliberation.json · Report an issue / request a re-scanAudit history
Every audit this server has had. A grade with a past is a grade somebody is still checking.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-07 | 5f4b6d348806 | BLOCK | F | 41 | first audit |
Questions
What is the Deliberation MCP server?
Ask Codex, Gemini, Grok, and 400+ OpenRouter models (Qwen, Kimi, DeepSeek) for second opinions or arbiter-mediated consensus. One MCP server for Claude Code, Codex, Cursor, Kiro, OpenCode. Measures the models' usefulness, too.
What tools does Deliberation expose?
18 in total: 11 read-only, 7 that write, and 0 that can delete or overwrite. Every one is listed on this page with its risk.
Is Deliberation safe to connect to an agent?
No — not without reading the findings first. The audit graded it F (41/100) and found 7 critical or high issues in the source. Each one is listed on this page with the file and line it is on.
What credentials does Deliberation need?
It reads API_KEY, FAKE_KEY and XAI_API_KEY from the environment. Give it a token scoped to the least it needs — an agent that can be talked into calling a tool can be talked into calling it with your credentials.
How does Deliberation run?
It speaks stdio, so it runs as a local process your client starts. It is published on npm as @antonbabenko/deliberation-mcp at 3.23.0.
How current is this page?
The grade is for one exact copy of the source (5f4b6d348806), read on 2026-10-07. The repository is watched and re-audited when it changes.