Skip to main content
Glama
r28ai

charter-packs-mcp

by r28ai

charter-packs-mcp

Charter packs, served over MCP: the runtime behind the R28 MCP servers, with their cross-app workflows and a login that keeps your own keys in your operating system's keychain.

Looking for the SDK? This is built on Charter, which turns an API endpoint into a typed tool any agent can call. Start there: pip install charter-ai.

A Charter pack is one API's tools. A family is the tools one kind of work calls across APIs: engineering is GitHub, Linear and Slack together, a support inbox is Gmail, Stripe and Linear. Each family carries the workflows that justify its tools, and its server offers those workflows as prompts (slash commands in Claude Code) beside exactly the tools they name.

Nothing here is part of the Charter SDK. This package is built on it: every tool is a pack tool, and the server is Charter's MCP adapter.

Nothing here is on PyPI: each family installs from a wheel on its own GitHub release, and depends on this package by the wheel on this repository's release. Only Charter comes from PyPI, and installing needs no git.

python -m charter_packs_mcp engineering                 # serve one family over stdio
python -m charter_packs_mcp support,finance             # two families, one server
python -m charter_packs_mcp engineering login           # connect the apps it uses
python -m charter_packs_mcp engineering status          # what is connected

Family

Workflows

Published as

engineering

25

github-linear-mcp

product

20

product-manager-mcp

sales

14

sales-prep-mcp

marketing

13

marketing-ops-mcp

support

12

support-inbox-mcp

finance

14

stripe-billing-ops-mcp

commerce

12

shopify-ops-mcp

ops

20

chief-of-staff-mcp

research

10

web-research-to-docs-mcp

people

8

recruiting-ops-mcp

agency

8

freelancer-ops-mcp

Connecting apps

Each server tells the agent at startup which apps are connected and how to connect the rest. Every credential is the user's own: nothing here is registered with any service, and there is no server of ours in the path.

  • Keys (GitHub, Linear, Slack, Stripe, Notion, Firecrawl, Tavily, Granola, Shopify) come from the client's config, or from login, which asks for them at a hidden prompt.

  • Google signs in through the user's own OAuth client, with PKCE and a loopback redirect, because a shared one would need Google's verification, and for Gmail and Drive a paid security assessment.

Whatever login or the connect tool stores goes to the OS keychain, after one read-only call to the app's own API confirms it works. A variable set in the client's config always wins over the keychain.

Related MCP server: AutoTool

Layout

src/charter_packs_mcp/
  catalogue.py    the 156 workflows: title, tools in order, why, cadence
  __init__.py     families, the MCP entry point, login / status / logout
  apps.py         each app: its key, the call that checks it, Google's sign-in server
  connections.py  where each app's credential came from, handed to the packs
  signin.py       `login`, Google's sign-in, the `connect` tool, the `setup` prompt
  keychain.py     the OS keychain, or an owner-only file without one
scripts/generate.py   writes ../charter-mcp-families: one publishable package per family

Development

uv sync
uv run pytest
uv run ruff check src tests scripts
uv run pyright
uv run python scripts/generate.py

Every test runs offline against respx. To work against a Charter checkout rather than the release, uv pip install -e ../charter after uv sync. Do not put that path in [tool.uv.sources]: uv honours it when the package is installed from GitHub, where ../charter does not exist, and every install fails.

License

Apache 2.0. See LICENSE.

Related MCP Connectors

Related MCP Servers

  • A
    license
    Not graded
    quality
    B
    maintenance
    Enables an agent to search approved MCP integrations and expose only the tools a given task needs, signing the user in once via browser-based OAuth and enforcing scopes, effects and budgets in code. It also acts as a single gateway connection so existing hosts can reach many servers with per-tenant isolation and audit events.
    MIT
  • A
    license
    Not graded
    quality
    A
    maintenance
    Connects AI clients such as Claude Code, Cursor, Codex, ChatGPT, Hermes, and OpenClaw to 2,000+ AI models and data APIs through a local MCP bridge, letting an agent discover, inspect, price, and run tools after a single browser-approved OAuth sign-in. Backed by a read-only catalog, run history, and rollback so calls can be audited and reverted without managing individual API keys.
    556,724 npm
    MIT
  • A
    license
    Not graded
    quality
    C
    maintenance
    Enables MCP clients to connect once to self-hosted, user-owned upstream accounts and then discover and invoke granted tools through a single authenticated endpoint. Each client receives its own key with deny-by-default tool grants, while sensitive calls require durable human approval and credentials stay encrypted and never forwarded upstream.
    Apache 2.0