council-mcp
Three-tier AI agent MCP orchestration system: Chancellor (Opus), Executor (Sonnet), Aide (Haiku)
Why this rank:Recent releaseHealthy release cadenceStrong adoption
Description
Three-tier AI agent MCP orchestration system: Chancellor (Opus), Executor (Sonnet), Aide (Haiku)
Release History
| Version | Changes | Urgency | Date |
|---|---|---|---|
| v0.8.0 | ### Added - **Numeric quality scores on Supervisor verdicts** â every Supervisor review now produces an optional integer `score` (0â100) alongside the boolean `approved`. Scoring rubric: correctness (40 pts) + completeness (30 pts) + intent alignment (30 pts). Scores are independent of the approval flag â a 75-point output can still pass if the task warrants it. - **`COUNCIL_MIN_SCORE` env var** â optional score gate (default: `0` = disabled). When set to a value 1â100, any Supervisor score bel | High | 5/18/2026 |
| v0.7.0 | ### Added - **Agent feedback loops** â closed two new inter-agent communication channels: - **Loop 2 (Executor â Chancellor coherence check)** â after all execution steps complete, the Chancellor is re-invoked in review mode (Haiku model, 1 turn, no file tools) to compare the original plan against the actual execution summary. Returns a structured `ChancellorCoherenceCheck` with `coherent` flag, `assessment`, `gaps`, and `recommendations`. Stored on the session as `coherence_check` and render | High | 5/13/2026 |
| v0.6.0 | ### Added - **Infrastructure retry with exponential backoff** (`withAgentRetry`) â every agent subprocess call is now wrapped in a 3-attempt retry layer with 1 s â 2 s â 4 s backoff (capped at 8 s). Only transient infrastructure errors (`AGENT_SDK_ERROR`, `AGENT_TIMEOUT`) trigger retries; parse/schema failures propagate immediately. Retry context (role, step) is logged at `warn` level on each attempt. - **Per-agent subprocess timeout** (`COUNCIL_AGENT_TIMEOUT_MS`) â hard deadline for each claud | High | 5/5/2026 |
| v0.5.0 | ### Added - **Supervisor evaluation loop** â the Supervisor is now an active quality gate, not advisory only. When it rejects an agent output (`approved: false`), the orchestrator re-invokes the agent with the flags and recommendation appended as feedback, up to `COUNCIL_EVAL_RETRIES` times (default: 2 â 3 total attempts). If the retry budget is exhausted and the output is still flagged, the result is surfaced anyway with the flags visible â no silent passes. - `COUNCIL_EVAL_RETRIES` env var â | High | 4/25/2026 |
| 0.4.0 | Imported from npm (0.4.0) | Low | 4/21/2026 |
| v0.4.0 | ### Added - **Per-agent tool access** â Chancellor and Aide now have tool access, configurable via `AGENT_TOOLS` constants: - Chancellor: `Read`, `Glob`, `Grep` (read-only â inspects codebase before planning, never writes) - Executor: `Read`, `Write`, `Edit`, `Bash`, `Glob`, `Grep` (unchanged) - Aide: `Read` (can read files before transforming them) - Supervisor: none (pure review, no side effects) - `AGENT_TOOLS` constant in `src/domain/constants/index.ts` â single source of truth for | High | 4/20/2026 |
| v0.4.0 | ### Added - **Per-agent tool access** â Chancellor and Aide now have tool access, configurable via `AGENT_TOOLS` constants: - Chancellor: `Read`, `Glob`, `Grep` (read-only â inspects codebase before planning, never writes) - Executor: `Read`, `Write`, `Edit`, `Bash`, `Glob`, `Grep` (unchanged) - Aide: `Read` (can read files before transforming them) - Supervisor: none (pure review, no side effects) - `AGENT_TOOLS` constant in `src/domain/constants/index.ts` â single source of truth for | High | 4/20/2026 |
| v0.4.0 | ### Added - **Per-agent tool access** â Chancellor and Aide now have tool access, configurable via `AGENT_TOOLS` constants: - Chancellor: `Read`, `Glob`, `Grep` (read-only â inspects codebase before planning, never writes) - Executor: `Read`, `Write`, `Edit`, `Bash`, `Glob`, `Grep` (unchanged) - Aide: `Read` (can read files before transforming them) - Supervisor: none (pure review, no side effects) - `AGENT_TOOLS` constant in `src/domain/constants/index.ts` â single source of truth for | High | 4/20/2026 |
| v0.4.0 | ### Added - **Per-agent tool access** â Chancellor and Aide now have tool access, configurable via `AGENT_TOOLS` constants: - Chancellor: `Read`, `Glob`, `Grep` (read-only â inspects codebase before planning, never writes) - Executor: `Read`, `Write`, `Edit`, `Bash`, `Glob`, `Grep` (unchanged) - Aide: `Read` (can read files before transforming them) - Supervisor: none (pure review, no side effects) - `AGENT_TOOLS` constant in `src/domain/constants/index.ts` â single source of truth for | High | 4/20/2026 |
| v0.3.0 | ### Added - **Persistent session memory** â sessions now survive MCP server restarts via an optional `COUNCIL_PERSIST` env var: - `memory` (default) â in-process LRU Map, no breaking change - `file` â JSON files at `~/.council/sessions/<id>.json`, zero dependencies - `sqlite` â SQLite at `~/.council/council.db` via `better-sqlite3`, WAL mode for safe concurrent access - **`SessionStore` interface** â all backends implement a common contract; swappable without touching orchestration code - | High | 4/17/2026 |
| v0.2.3 | ### Changed - README diagrams replaced with hosted images â architecture, orchestration flow, and session lifecycle diagrams now render correctly on all platforms including GitHub, npm, and PyPI mirrors --- **Install:** `npm install council-mcp` **Use in Claude Code MCP config:** ```json { "mcpServers": { "the-council": { "command": "npx", "args": ["-y", "council-mcp"] } } } ``` | Medium | 4/12/2026 |
| v0.2.2 | ### Fixed - `install.sh` now also runs `claude mcp add` to register the server with Claude Code CLI â previously only Claude Desktop was configured, so the tools were invisible in the CLI - `install.ps1` rewritten to match: detects `claude` binary, falls back to `ANTHROPIC_API_KEY`, configures both Claude Desktop and Claude Code CLI - `runner.ts` strips `ANTHROPIC_API_KEY` when set to an empty string â Claude Desktop injects an empty key into the MCP server env, causing the child `claude` proce | Medium | 4/12/2026 |
| v0.2.1 | ### Fixed - Replace `@anthropic-ai/claude-agent-sdk` with direct `claude` CLI subprocess calls â eliminates 401 auth errors for users authenticated via Claude.ai OAuth (no separate API key needed) - Use `--system-prompt-file` instead of `--system-prompt` CLI arg â prevents `exit 1` failures caused by long system prompts with XML tags and special characters - Startup check fails fast with a clear message if `claude` CLI is not in PATH and no `ANTHROPIC_API_KEY` is set - `install.sh` and `install | Medium | 4/12/2026 |
| v0.2.0 | ### Added - **Supervisor agent** (Claude Haiku 4.5) â reviews every Executor step result and Aide task output before they surface to the caller. Non-blocking: if the Supervisor errors, orchestration continues and a warning is logged. - **`get_supervisor_verdicts` MCP tool** â retrieve all Supervisor verdicts for a session, with optional `flagged_only` filter for quick triage - **Supervisor flags in result summary** â flagged outputs appear under a `## Supervisor Flags` section in the `orchestra | Medium | 4/11/2026 |
| v0.1.2 | ### Fixed - Release workflow now reconfigures npm registry to `npm.pkg.github.com` before publishing to GitHub Packages, fixing `ENEEDAUTH` on the GitHub Packages publish step --- **Install:** `npm install council-mcp` **Use in Claude Code MCP config:** ```json { "mcpServers": { "the-council": { "command": "npx", "args": ["-y", "council-mcp"] } } } ``` | Medium | 4/11/2026 |
| v0.1.0 | ### Added - **MCP server** with five tools: `orchestrate`, `consult_chancellor`, `execute_with_executor`, `delegate_to_aide`, `get_council_state` - **Chancellor agent** (Claude Opus 4.6) â strategic analysis, risk assessment, and step-by-step planning via the Agent SDK - **Executor agent** (Claude Sonnet 4.6) â plan implementation with access to `Read`, `Write`, `Edit`, `Bash`, `Glob`, `Grep` tools - **Aide agent** (Claude Haiku 4.5) â simple tasks: formatting, data transformation, utilities - | Medium | 4/11/2026 |
Dependencies & License Audit
Loading dependencies...
Similar Packages
automagik-genieSelf-evolving AI agent orchestration framework with Model Context Protocol supportv4.260606.2
@claude-flow/cliRuflo CLI - Enterprise AI agent orchestration with 60+ specialized agents, swarm coordination, MCP server, self-learning hooks, and vector memory for Claude Codev3.10.34
agent-link-mcpMCP server for bidirectional AI agent collaboration â spawn and communicate with any AI coding agent CLI0.5.0
More in MCP Servers
claude-plugins-officialOfficial, Anthropic-managed directory of high quality Claude Code Plugins.
langchain4jLangChain4j is an open-source Java library that simplifies the integration of LLMs into Java applications through a unified API, providing access to popular LLMs and vector databases. It makes impleme
hyperframesWrite HTML. Render video. Built for agents.
claude-code-guideClaude Code Guide - Setup, Commands, workflows, agents, skills & tips-n-tricks go from beginner to power user!