Claude and agents

The PYLO plugin for Claude

Install PYLO into Claude Code and run onboarding, weekly reviews and integration fixes from a conversation.

Read as Markdown

What it is

The PYLO plugin connects Claude to your workspace through PYLO's MCP server and adds three skills: step by step playbooks Claude follows when you ask for the job they cover. Claude reads your figures, proposes changes as plans a person approves, and never sees a credential.

The plugin comes in two versions with the same skills and the same tools:

Install one of the two, not both. They expose the same tools.

Install

You need Claude Code and a PYLO token. A workspace token from API tokens is read only, which is enough for weekly reviews and for diagnosing an integration. Planning and applying changes, sending connect links and running test leads need an agency token from Agency settings, issued with those permissions. Either is shown once, so copy it straight into Claude Code when asked.

In Claude Code, add PYLO's plugin marketplace, which is the public repository chigozieumeukeje19-netizen/pylo-claude-plugin, then install the plugin:

/plugin marketplace add chigozieumeukeje19-netizen/pylo-claude-plugin
/plugin install pylo@pylo

Claude Code then asks for two settings:

To sign in with your PYLO account instead, install the second plugin. It asks only for the MCP URL, then opens PYLO's consent screen on first use, where you choose what Claude may do. PYLO caps what you grant to what your role already allows.

/plugin install pylo-oauth@pylo

Just the tools, without the skills

If you only want Claude to read PYLO, add the MCP server on its own:

claude mcp add --transport http pylo https://api.pylosystem.com/mcp --header "Authorization: Bearer YOUR-TOKEN"

Every tool it offers is listed in the MCP reference.

The skills

weekly-review

Ask for a weekly review, a health check, or where a client's funnel is leaking. Claude changes nothing. It first asks PYLO whether the week's figures can be trusted, then finds the one stage losing the most, proves it from PYLO's own figures, and proposes at most two actions.

It follows fixed rules. Every figure comes from PYLO and is never estimated. Every rate is stated with both sides, for example a show rate as live calls out of live calls plus no-shows. A figure PYLO cannot compute is reported with its reason, never as zero. Money keeps the currency PYLO returned.

fix-integration

Ask why a tool stopped syncing, why data is missing, or why notifications stopped. Claude reads your connector health and the data trust checks, works out why a provider is quiet or failing, and proposes the fix as a plan for a person to approve or a connect link for the client to complete.

It never asks for an API key, token, webhook address or password. If one is pasted into the conversation, it tells you to revoke it. It only calls something fixed once PYLO reads it back as working.

onboard-client

Ask Claude to onboard, set up or launch a client. It runs the whole onboarding from the conversation: it drafts the workspace from a template and the client's intake, shows the plan, waits for a person's approval, applies it, sends the client one connect link for their tools, plans again once they are connected, and runs a test lead through end to end.

Planning a brand new workspace needs an agency token. An owner signed in through pylo-oauth works on their own workspace.

What Claude can and cannot do

Claude can read anything the token or your sign-in allows, and it can plan and apply configuration once a person approves. It cannot move money, refund, pay out, edit a payment or delete anything: PYLO has no tool for any of those. Changes are always shown as a plan first and applied only with a person's approval.