claude-council

Claude Code plugin that asks several AI coding agents the same question and shows their answers side by side. Gemini, OpenAI, Grok, Perplexity, Kimi and any model OpenRouter routes to over API; codex, antigravity, grok, kimi and cursor CLIs on subscription auth; ollama locally. Codex specialists take coding tasks on th

Automatically checked · GitHub checked 2026-11-14

claude-council

Overview

claude-council

A Claude Code plugin that asks AI coding agents the same question in parallel. Compare their answers when one model's bias could mislead you: architecture decisions, debugging dead ends, security reviews, or framework choices.

Codex, Antigravity, Gemini, and OpenAI answer the SQLite-or-Postgres question while fake Kimi fails

Tests

Where it runs

WhereWhat you get
Claude Code in a terminal/claude-council:ask, every seat, and the pane beside the chat.
Claude Desktop, Code tabThe same plugin and the same commands. The pane opens in Desktop's side panel.
Claude Desktop, an ordinary chatThe Desktop extension: five tools, a seat picker and an answers panel in the chat. macOS and Linux.

Claude Desktop's Code tab: Claude's synthesis of a primary-key question on the left, the Council pane with each seat's answer on the right

In both, Claude can also call the council as a tool. It asks which seats and how verbose first, then shows a Send card, and nothing leaves your machine until you confirm it.

claude.ai chat and Cowork run in a sandbox that can block calls to other providers, so the council does not work there.

Install

You need curl and jq on macOS, Linux, or Windows through Git Bash.

/plugin marketplace add hex/claude-marketplace
/plugin install claude-council

Give it at least one provider in the shell you start Claude Code from:

export OPENAI_API_KEY="..."
# Or GEMINI_API_KEY, XAI_API_KEY, PERPLEXITY_API_KEY,
# KIMI_API_KEY, OPENROUTER_API_KEY

Or store the keys in the plugin's config. Run /plugin configure claude-council, fill in the API key rows, and choose Save configuration. Claude Code keeps them in its secure storage, not in a settings file. The council's mod sets each stored key as its variable for the whole session, so every command Claude Code starts, in every project, gets the stored key in place of one your shell exports. The OpenRouter models row takes the model ids OPENROUTER_MODELS takes, so the router's seats need no export either. Stored keys need Claude Code 2.1.287 or later. The desktop app's plugin Settings tab takes them too, for sessions on that computer. They never reach claude.ai chat or Cowork. API keys covers where they are kept, when the mod does not load, and how a cleared key behaves.

The codex, agy (Antigravity), grok, kimi, and cursor-agent CLIs use your existing subscriptions without an API key. Claude Code itself, the Copilot CLI and OpenCode can sit too, when named in --providers. A local ollama model needs no key or external network. With none of these, --local uses Claude subagents. See provider selection and local councils.

First run in 60 seconds

Read what gets sent and saved before asking about your code.

/claude-council:status
/claude-council:ask "Should I use UUID or BIGINT primary keys for a SaaS users table?"

Each provider answers independently. Claude writes a synthesis of their agreement, disagreements, and recommendation. Agreement is not verification: the models share your description and its assumptions. The synthesis names the premise that could change the answer.

Claude Desktop

The plugin needs Claude Code. For an ordinary Claude Desktop chat there is a Desktop extension, council-<version>.mcpb, that gives Claude five tools: council_ask, council_pick, council_result, council_status and council_settings. It runs on macOS and Linux only, because it starts the council's bash scripts.

  1. Download council-<version>.mcpb from the latest release. To build it yourself, run bash extension/build.sh in a clean checkout. It needs npm and writes dist/council-<version>.mcpb.
  2. Double-click the file. Desktop warns that the extension gets access to everything on your computer and is not verified by Anthropic. Both hold for any extension installed from a file: its server runs as you, with your files and network, and Anthropic has not reviewed it. Install only a package you built or trust.
  3. In Settings, Extensions, Claude Council, Configure, paste a key for each provider you want: Gemini, OpenAI, xAI, Perplexity, Kimi, OpenRouter. An empty field leaves that provider's API out. Desktop stores the keys. For a question, each key goes only to its own vendor's seats: the vendor's API, and its CLI agent when one is installed (codex gets the OpenAI key, agy the Gemini key, grok the xAI key, kimi the Kimi key). A status check is different: it starts each installed CLI agent but claude and opencode with every key set.
  4. In a chat, ask Claude to open the council settings. A panel in the chat lists each keyed provider's models, fetched live from that provider. Pick one per provider, or switch on up to 12 OpenRouter models, one seat each. Every change saves at once and applies to the next question.
  5. Ask Claude to put a question to the council. Desktop asks you to approve the tool call, and that approval is the only check before the question leaves your machine. A panel under the tool call shows each seat's answer, or its error, in a card of its own headed by the seat's name and its vendor's logo, and Claude writes the synthesis below it. Desktop waits for the whole run, from seconds to a few minutes.

Asking at once

Name the seats or a verbosity (standard, brief or detailed) in your message, or say to just ask everyone, and Claude runs the question at once, on every default seat at standard unless you named others. Claude passes your question as you asked it, adding context only when it depends on earlier messages the council cannot see.

The picker

Otherwise, as in "ask the council: which queue should we use?", Claude opens council_pick, a panel with the question, a switch per seat and the three verbosities. Claude then waits, in the same turn, in council_result for the picker's run.

  1. Press Ask. The answers appear in that panel, and the line under them says "Claude is writing the synthesis".
  2. Claude writes the synthesis with no message from you. When every seat fails, Claude says what failed.

When no seat answers, the panel says so and Ask works again.

Changing a model in the picker

On an API seat or an OpenRouter seat the model in grey is a button: click it to pick another model from that provider's live list (OpenRouter's has a filter box). The choice saves at once, as the settings panel's would, and applies to this question and the ones after it. An OpenRouter seat changes only its own model; adding or removing OpenRouter seats stays in the settings panel.

When Claude is no longer waiting

Claude waits at most 11.5 minutes from when the picker opened. If it has stopped waiting by the time the answers are in, or you press Ask a second time in the same panel:

  • Desktop puts "Synthesise the council's answers (run 3f9a2c)." in the message box, under a caution about malicious content, and the line under the answers says to press Enter. The panel cannot send that message itself.
  • Claude then fetches that run's answers with council_result and writes the synthesis.
  • If Desktop refuses the message, the line instead gives the words to send yourself (Ask Claude: "synthesise council run 3f9a2c").
  • The extension keeps the last 5 runs in which a seat answered; after that, ask the council again.

A question with no providers named goes to every provider with a key and to every CLI agent the scripts find (codex, agy, grok, kimi, cursor-agent, ollama) in your PATH, ~/.local/bin, ~/bin, /opt/homebrew/bin or /usr/local/bin. claude, copilot and opencode get it only when the question names them. A CLI agent answers on its own login, so it receives the question even when its vendor's key field is empty; apart from its own vendor's key it gets no key from the form. The reference lists the files it keeps and its limits.

Features

CapabilityGuide
Ask providers together, or name one in plain wordsCommand examples
Choose providers and pin a standing rosterProvider selection
Assign specialist rolesRoles
Have providers challenge each other's answersDebate
Gather code context with agents before askingAgent mode
Use Claude subagents without an external providerLocal council
Include relevant files or choose your ownAuto-context
Attach a screenshotImage input
Ask about the current conversation/advise
Run in the background and fetch the resultBackground jobs
Export the answers to a fileExport
Show only the synthesisQuiet mode
Watch answers arrive and retry failed seatsThe pane
Get a note from one seat while Claude worksHeads-up side-check
Have a second model review the diff before Claude stopsStop-gate review
Ask the council from a Claude Desktop chatClaude Desktop
Call the scripts without Claude CodeDirect script usage

The reference lists every flag, variable, model default, and limit. Architecture covers the implementation. TESTING.md covers tests, releases, and adding a provider.

Heads-up side-check

The intro hint, a heads-up note, Learn more, and an unsent Tell Claude draft

Opt in through the "Heads-up side-check" row in /config or the project's .claude/council-heads-up.json. One seat, codex by default, checks recent turns and tool results while Claude works. For a Write or an Edit, it gets the text Claude wrote. The note shows what went where. Learn more expands it; Tell Claude puts a draft in the prompt without sending it. Read the data limits and off switch and the heads-up guide.

The pane

On Claude Code 2.1.287 or later, the pane opens inside Claude Code, as in the animation above. Each seat has one row with its model, a bar for its state, its time, and a button for what it does now: cancel, retry, open, or fold. When the run ends, the synthesis opens at the top with the answers folded below it. Reopen the last run with /council-pane. Inside tmux, the mod asks once which pane you want.

The mod guide lists keys, settings, limits, and when the mod does not load. It also lets Claude call the council as a tool; every call asks you first.

What the mod runs, reads and sends

The mod makes no network calls of its own. It starts the programs below from an argument array, never a shell string, and never hands bash -c a script. Your text reaches model providers only through the first two:

  • bash scripts/run-council.sh when Claude calls the council tool and you choose Send to the council. It sends your question to the providers you set up.
  • bash scripts/heads-up.sh, only with the heads-up side-check on. It reads this session's transcript (by default the last 5 turns and up to 8 tool results of 2,000 characters each) and sends that digest to one seat, codex unless you pick another. The run directory keeps digest.md and prompt.md, exactly what left the machine, until the mod reads the result and removes it. The codex seat runs in that directory under a codex home of its own: your model and effort, a link to your auth.json so a token refresh reaches your login (without one, your CODEX_API_KEY or OPENAI_API_KEY signs codex in), and none of your plugins, hooks, AGENTS.md, skills or MCP servers.
  • bash scripts/query-council.sh --list-default-models when Claude calls the council tool, to list the seats it may offer you and to check the seats it names. It reads your configuration and sends nothing.
  • bash scripts/seat-info.sh discover, model <seat> or allowlisted <seat> during /heads-up setup and /heads-up on <seat>. It checks which seats this machine can run and sends nothing.
  • codex debug models during /heads-up setup with the codex seat, to list its models. That command belongs to codex, and so does any request it makes.
  • kill -0 <pid> to check a council run is still alive. It sends no signal.
  • rm -rf and mv -f on its own run directories under $TMPDIR/council-mod.<session id>, to clean up a finished run and to answer the retry offer.

It writes files in that $TMPDIR directory, including model.json and avoid.txt, which heads-up.sh reads. Outside it, it writes one file, .claude/council-heads-up.json in the project, and only from /heads-up setup, /heads-up on, /heads-up off or a note's Turn off. That file is the heads-up switch. Claude Code itself does not read it.

It sets environment variables in Claude Code's own process. With an API key or the OpenRouter models list stored in the plugin's config, it sets that variable, such as OPENAI_API_KEY or OPENROUTER_MODELS, when the session starts. Every command Claude Code starts after that gets the stored value in place of the one your shell exports, the same as an export would. Before a council run it sets COUNCIL_MOD_PANE_DIR.

It touches the prompt in two places. With wake_on_async_done on, it submits The background council job <id> has finished. Fetch it with /claude-council:result <id> and summarise it. once a background job finishes. A note's Tell Claude puts a draft in the prompt box, and you decide whether to send it.

It reads the text of each Bash command Claude runs, only to spot a call to run-council.sh. Just before that call it sets COUNCIL_MOD_PANE_DIR, so the run draws in this pane, and clears it when tmux owns the pane. It reads TMPDIR and TMUX. It turns on the claude-council.heads_up setting only when you pick "every project" in /heads-up setup.

Real SQLite-or-Postgres answers arrive in tmux beside a fake Kimi failure and the retry offer

Without the mod, tmux opens a side pane that streams answers and each provider's status. It offers to retry failed seats. Esc closes it. --no-pane or COUNCIL_NO_PANE=1 turns it off. Outside both pane modes, answers arrive in the chat. See terminal integration.

The council and tmux animations show real answers to "Should a single-node demo app use SQLite or Postgres?" Codex and Antigravity use subscriptions; Gemini and OpenAI use their APIs. Kimi is a rate-limit fixture. The heads-up animation uses a fake seat and a real Claude Code host. Capture scripts and raw frames record how to reproduce them.

What a CLI seat loads

Every CLI seat runs from an empty directory, $TMPDIR/claude-council-<uid>/seat, where <uid> is your numeric user id. The CLAUDE.md, AGENTS.md, rules, skills, hooks and MCP servers of the project you ask from don't reach it. The directory doesn't limit what a seat can read: codex, for example, can still open any file by its full path.

The directory is read-only. That stops a seat writing a file there, but not a seat that changes the mode first. A seat also refuses to start when it finds anything in it, and names the path so you can look before you remove it. A file that lands while other seats are running can reach them; the next seat to start refuses. A seat also refuses to start when TMPDIR sits inside a git repository, and names it: a CLI would take the whole repository as the seat's project.

The seats below also run under a home of their own, in $TMPDIR/claude-council-<uid>/seat-home. It links only to what the CLI needs to sign in and start. Only you can open either directory, and a seat refuses to start when one of them belongs to another user or is a symlink.

SeatNo longer loadsStill loads
cursor-cliSkills and plugins from ~/.cursor, ~/.claude, ~/.codex, ~/.grok and ~/.agents, Cursor and Claude hooks, ~/.cursor/mcp.jsonTeam commands from your Cursor dashboard (read from the CLI's bundle, not seen in a run)
grok-cliClaude plugins, ~/.agents skills, grok's memory across sessionsYour ~/.grok config and the hooks in ~/.grok/hooks, since no grok setting turns them off. The seat home links your login keychain (macOS) and ~/.docker/run, for Docker Desktop's socket
antigravityOn macOS: your ~/.gemini rules, skills, MCP servers and saved conversationsOn Linux: everything under ~/.gemini, because where agy keeps its login there is not confirmed
opencodeThe ~/.claude and ~/.agents skills, agents and CLAUDE.mdYour login and session history in ~/.local/share/opencode
codexYour plugins, hooks, skills, memories and MCP servers. Each question gets its own codex home with your model, your effort and a link to auth.jsonNothing. With no auth.json and cli_auth_credentials_store set to keyring or auto, your login is in the system keyring, which codex can't read from another home, so the seat runs under your own codex home and the answer starts with a note. An exported OPENAI_API_KEY doesn't change that; an exported CODEX_API_KEY still signs codex in ahead of any login, as it does when you run codex yourself. Set cli_auth_credentials_store = "file" and run codex login again to isolate it. With no login at all, the seat stops and says so
claude-cliThe git status and commit instructions a repository addsAdmin policy and your organisation's instructions
kimi-cli, copilot-cliThe project's filesYour home: kimi's hooks in ~/.kimi-code/config.toml, copilot's skills in ~/.copilot and ~/.agents

Cursor and Antigravity keep the model you picked in their own config: the seat reads it from your real home and passes it on. On Windows the seats keep your home, because the native CLIs find it through USERPROFILE. Seat isolation has each seat's details.

Troubleshooting

ProblemFix
Cloned the repo and nothing loadsDon't clone into ~/.claude/plugins/. pluginDirectories is not a setting. Follow running from a local clone.
No providers foundRun /claude-council:status. Export keys in the shell that starts Claude Code, store them with /plugin configure claude-council, or use --local.
A stored key is not usedAfter saving one with claude plugin configure in a shell, restart Claude Code. Stored keys need the council's mod, which loads on Claude Code 2.1.287 or later and not under --safe-mode or disableAllHooks. Without it, export the key in your shell.
A different model answeredA header such as grok-4.7 (grok-latest unavailable) records a model fallback. Setting that provider's _MODEL variable disables fallback.
No paneUse tmux, or update to Claude Code 2.1.287 or later for the pane inside Claude Code. The mod guide lists what stops it loading.
UpdatingRun /plugin update claude-council, then restart Claude Code.