Brew Context · Context for AI coding agents

Your agents get one tap. You control every pour.

on tap: controlled tool access|

Brewale connects Claude Code and any MCP-compatible agent to your MCP servers through a single connection — with per-tool access control, read/write/delete classification, and one-click revocation. Then serve the good stuff: versioned skills and coding conventions, shared across your whole team.

Tap in Read the brew sheet

Start free — 14-day trial, no card. The clock only starts on your org's first MCP request.

The whole brewery, at a glance

Skills

Versioned markdown bundles, served on draft.

Conventions

One document, one source of truth for every agent.

Tool access

Per-tool ACLs on every MCP server you bring.

https://mcp.brewale.dev/mcp

One governed connection to Claude Code, Codex CLI, Cursor, VS Code, Zed — any MCP-compatible agent.

Right now, every agent drinks straight from the keg

Connect an MCP server to an agent and the agent gets every tool it exposes — delete endpoints sitting next to read endpoints, no distinction between them.

Credentials get pasted into config files, one per agent, per laptop. Offboard someone and you're hunting tokens.

There's no audit of who can call what, because “who can call what” was never a setting — it was whatever ended up in .mcp.json.

That's fine for a demo. It's not fine for a team.

Pour control

Per-tool ACLs on every MCP server you bring

Bring your own MCP server — bearer token, OAuth with Dynamic Client Registration, or no auth at all. Brewale discovers its tools automatically on publish and proxies them to your agents through the single Brewale connection. Then govern the pour:

Toggle each tool on or off — nothing is exposed by default accident.
Classify every tool as read, write, or delete.
Set org-wide defaults with per-tool overrides, or scope ACLs per user — and enforce permission ceilings nobody can exceed.
Credentials are shared org-wide or held per member (OAuth is always per member), and every token is stored encrypted at rest.
Revoke in a click — it takes effect immediately.
Draft → publish lifecycle: nothing reaches an agent until you say so.

From guest tap to governed pour

Three steps to put every tool under pour control

Connect

Add any MCP server as a guest tap: bearer token, OAuth via Dynamic Client Registration, or no auth at all.

Classify

Brewale discovers its tools; you toggle each one, mark it read, write, or delete, and set defaults, overrides, and ceilings — org-wide or per user.

Pour

Publish. Agents reach every allowed tool through one Brewale connection. Revoke access in a click, effective immediately.

Skills on draft

What you serve once the taps are governed

Governance is what makes sharing safe. With the taps under control, pour your team's actual knowledge: skills and a conventions document, written in plain markdown, versioned with full history, and served to every agent over MCP — get_conventions, list_skills, get_skill, get_skill_file.

Agents always receive exactly the published version — immutable content, one active pointer, unpublish supported — and a published change reaches every connected agent within about 5 seconds. Skills ship as bundles when they need to: SKILL.md plus references, scripts, and templates. One source of truth instead of per-repo CLAUDE.md drift.

A shared library, a personal glass: each member toggles individual skills for their own sessions — and can keep private personal skills only their own agents ever see.

The library grows itself

create_skill agent proposes from a live session → human reviews in the app → published straight to your repo

Skills-as-code

Your skills. Your files. GitHub optional.

Author skills straight from your agent over MCP — no repo required — and download everything as markdown whenever you like. Prefer Git? Install the Brewale GitHub App and your conventions.md and skills/ directory sync on every push to the default branch — reviewed like code, because they are code: plain markdown you own outright. Leave anytime and keep every file.

It's your knowledge; Brewale is just the tap.

Projects — pour by the codebase

One org, many repos, different standards. Scope which skills and conventions each agent receives per project, with an always-served set on top. Agents register and find projects over MCP — register_project, search_projects — so the right context flows to the right codebase. Enabled per org in Settings → Organization.

What's pouring

Per-skill analytics show total calls, the 30-day trend, and top callers. Per-member analytics track skill and tool calls, so you can see what's earning its keg and where the gaps are.

Under lock and keg

Identity for humans, agents, and machines.

OAuth 2.1

Dynamic client registration and a browser consent flow in supported clients like Claude Code — no token copy-paste.

API keys

Org-scoped Bearer tokens for clients where OAuth isn't an option.

Service accounts

Headless agents and CI get their own identity — OAuth or bearer token.

Two-factor auth

TOTP for every human account, plus Sign in with Google with account linking.

Instant revocation

Self-service, effective immediately. Mark a member inactive and their MCP access pauses — and they're not billed as a seat.

First pour in one command

Works with any MCP-compatible agent over the Model Context Protocol. No SDK, no glue code.

$ npx brewale

Claude CodeCodex CLICursorVS CodeZed

Prefer to keep the repo clean? The --local scope for Claude Code writes nothing to the repo — no committed .mcp.json.

Ready to take pour control?

Every agent pours from the same tap — the same skills, the same conventions, and only the tools you've approved.

Simple, per-seat pricing. Everything included. 14-day free trial, no card required — and the trial doesn't start until your org's first MCP request.

Set up untimed

First MCP request clock starts

Day 1 of 14 free trial

Start your first batch Read the brew sheet