MSP Skills: Query PSA, RMM, Backup, and M365 Data in Claude Code with Natural Language

Who it's for

This Skill collection is for MSP (managed service provider) engineers who manage multiple client lines: your daily work revolves around PSA, RMM, backup, and M365, and you would rather have an AI assistant look things up for you in natural language than log into each backend one by one. It is especially suitable for teams serving many clients at once who need to compare system states quickly; for a single client with only an occasional one-off lookup, the benefit is limited. The official positioning is a free collection of MCP servers and Claude Code Skills with a local-first emphasis, and it claims "your data never leaves your network" — that is the vendor's statement, and the actual boundary depends on your configuration and usage.

When it is not a fit: you cannot obtain access credentials for the systems in question (the prerequisite is supplying your own credentials at runtime, and the vendor explicitly states that the repo stores nothing); you want your data hosted and processed by a third-party cloud; your environment requires all data to be retained centrally or forbids local database files; or your AI assistant supports neither Claude Code plugins nor MCP. Which products and versions each connector supports is not listed in the official documentation and remains unconfirmed — check the repo before choosing.

First try in 3 minutes

There is exactly one recommended main path: install through the plugin marketplace inside a working Claude Code. Prerequisites: you have a login account for one of the systems you want to query (one of PSA, RMM, backup, or M365), and credentials must be supplied by you at runtime. Run the following in a session:

/plugin marketplace add https://github.com/servosity/msp-skills.git
/plugin install connect-tool@msp-skills

Once both commands finish, the Skill is installed. Next, connect and sync a system, which requires providing that system's credentials; the exact configuration commands are not given in the official documentation and remain unconfirmed — from this point on, follow the repo's current page, and before entering any credentials, confirm that you are interacting with this tool.

After connecting one system, a good first task is a read-only lookup: ask the AI assistant in natural language to retrieve one piece of information from the local mirror, such as summarizing a client's recent backup status. Queries issued before syncing completes will lack data sources; that is normal. In the official workflow, every connected system syncs into a local SQLite mirror with full-text search, so cross-client and cross-engine questions become a single local query: instant, offline, and the AI sees the answer rather than the raw data. Success signals: the answer comes from the local mirror instead of live calls to external APIs; if you see errors about missing connections or credentials, finish connecting first and then retry.

The vendor also mentions that each installer drops both the CLI and the MCP server: the Skill form is used in Claude Code and Codex, while the server form serves clients such as Claude Desktop, ChatGPT, and Cursor; the exact installer commands are not given in the official documentation and are not covered here.

Use cases

Case 1: unified cross-client search. Scenario: investigating cross-client questions such as "which clients have abnormal backup status", which used to mean logging into each backend one by one. How to start: in Claude Code, ask the AI assistant in natural language, naming the clients and objects to query — what actually runs is a single full-text search query against the local SQLite mirror. What to expect: per the official description, the query is instant and can work offline, and the AI gets the answer rather than the raw data, provided the relevant systems are connected and synced. Naming the client and the query target clearly makes the answer more focused; if a result disagrees with the backend's actual data, re-check with the backend as the reference.

Case 2: one install, reused across clients. Scenario: a colleague prefers clients like Claude Desktop or Cursor and does not want to install and configure everything again. How to start: the single installation already provides both the CLI and the MCP server, so connect in an MCP-capable client and ask the same question. Result: the same local-mirror workflow still works from another client. The specific connection steps for each client are not given in the official documentation; what can be confirmed ends at "one install provides both forms".

Usage boundaries

• Credentials: supplied by you at runtime; the vendor explicitly states "Nothing is stored in this repo". Credentials are sensitive — enter them only in the official tool's interface, and verify what you are interacting with before typing.
• Local and network: syncing writes the connected systems' data into a local SQLite mirror — this is the officially stated storage method — and syncing uses your credentials to access the corresponding online systems; the mirror file stays on your machine, but its exact location is not given in the official documentation and remains unconfirmed. "Your data never leaves your network" is a vendor claim without independent verification and should not be treated as a security conclusion.
• Client prerequisites: the vendor lists Claude, ChatGPT, Codex, Cursor, Windsurf, and any AI assistant that speaks MCP; the list does not mean other clients are unusable — whether unlisted clients work is unconfirmed, so confirm your client's connection form before choosing.
• Unverified information: the connector list, the products and versions each tool supports, the exact commands for credential configuration and installers, and the per-client connection steps are all absent from the official documentation; before use, follow the repo's current page.