Subscriby Agentic Kit
A Claude Code plugin that teaches Claude to run a community, integrate the API and webhooks, and build a connector on Subscriby, with the MCP server bundled and every destructive tool guarded.
The Agentic Kit is what you install once in Claude Code and keep: the knowledge of how to run a community, integrate an application or build a connector on Subscriby, packaged as a plugin. It bundles the hosted MCP server, so Claude signs in as you; it carries a skill for every job, slash commands for the routines, read-only subagents for audits and reports, a hook that makes every destructive tool ask before it runs, and an eval suite that proves each release before it ships.
/plugin marketplace add envigoinnovations/subscriby-agentic-kit
/plugin install subscriby@envigoThe kit installs 36 skills in four categories (3 for the foundation, 17 for creators, 10 for developers, 6 for connector authors), 8 slash commands, 3 read-only subagents, a companion guard plugin whose hook makes every one of the 84 destructive tools ask before it runs, and an eval suite. Every tool name, event, ability and count in it is generated from the catalogue Subscriby exported on 2026-05-01; this page describes kit 1.1.0.
Who it is for
Creators
Run the community from a conversation: launch a project, read the week's figures, preview and send a broadcast, triage the inbox, issue access codes, replace a lost place. Every write is confirmed, every figure comes with the tool call behind it.
Developers
Integrate Subscriby into whatever you are building: a verified webhook receiver written into your repository, the events a job needs, the least set of abilities for a token, a client that retries the right way, the Zapier and n8n nodes, and an agent of your own.
Connector authors
Build a connector on the SDK with the manifest, the ports, the data rules, the fakes and the conformance kit at hand, from an empty package to a submission.
What is inside
Skills
36 skills in four groups, each a concise procedure with the concept model, the confirmation points and a generated reference of the tools, events, endpoints, abilities or error codes it owns. Standard SKILL.md files, so they load wherever skills do.
Slash commands
8 commands you type for the routines: sign in, a status card, a project's doctor, the guided launch, the week in figures, a documentation lookup, a webhook verifier for your repository, the least token for a purpose.
Subagents
3 read-only agents on generated tool allowlists: a launch auditor, a community analyst and an integration reviewer. They report; they never change anything.
Safety
A companion guard plugin that makes every tool flagged destructive ask before it runs, whatever your permission settings say, and the rules every skill follows: read first, confirm every change, never a token in the conversation.
Evals
A suite of cases phrased as users type them, graded on the answer and on the steps taken against mocked tools with the real schemas, run before each release and compared with a no-plugin baseline.
Never more than your own permissions
The kit adds knowledge, not reach. The bundled server signs in as you, so Claude can do exactly what you can do on the team you chose at consent and nothing more; a personal access token with fewer abilities narrows that further. Every tool enforces its own ability on the server, the same one its REST twin requires, and the activity log records each call with the actor behind it.
Questions
Related
- MCP server: the server the kit bundles, and how every other client connects to it.
- Building custom agents: an agent of your own on the REST API and webhooks.
- Connector SDK: the reference the connector skills rest on.
How is this guide?
Last updated on
Postman / Insomnia
Import the ready-made Postman collection, or the OpenAPI document, to explore every endpoint with a personal access token and test a webhook receiver against signed deliveries.
Install and Sign In
Add the envigo marketplace, install the subscriby plugin and its guard, sign in to the bundled MCP server and prove the connection, in the terminal, the VS Code extension, claude.ai or any other agent.