MCPulse skills

Reusable skills for AI agents that work with MCPulse, analytics for MCP servers.

getmcpulse.com · Docs · Dashboard

A skill gives your agent a versioned workflow guide instead of leaving it to infer how MCPulse fits together from tool descriptions alone. A description says what one tool does; it cannot say which one to reach for, or which figures look wrong but aren't.

Install

npx skills add getmcpulse/skills --skill mcpulse

Restart your AI tool if it doesn't pick the skill up straight away.

Pair it with the MCP connector — the skill explains the workflows, the connector executes them.

Skills

mcpulse

Reading and diagnosing MCP server analytics over the MCPulse connector:

• the account → MCP → tool / session / call shape, and why a tool exists from the moment a client initialises rather than from its first call
• narrowing get_overview with metrics instead of pulling the whole overview to read one field
• naming an MCP directly rather than spending a list_mcps round trip, and what to do when a name is unknown or ambiguous
• the diagnostic order — insights, then rank by calls × failure rate, then the outcome breakdown, which separates a schema problem from a silent-empties problem from a description problem
• which figures are already computed server-side and must not be re-derived: deltas, direction, cost, insight verdicts
• addressing an MCP by name, a member by email and a key by prefix, and the writes with a trap in them — update_notifications replaces rather than merges, create_mcp returns no key
• the three metrics that lag a day, and saying so when quoting them
• the difference between a null that means "not yet computed" and a 0 that means "nothing happened"

It also names the behaviours that read as faults and are not: blanked sessions under a tool filter, a p95 that is a bucket edge rather than a value, empty tool pairs, insights suppressed below 20 calls, and a "dead" tool that is simply outside the date range.

Why a skill

Without one, an agent works MCPulse out from tool descriptions each time. With one it gets a stable reference for how the product fits together, use-case driven guidance rather than tool-by-tool guessing, explicit caveats for what the connector cannot do, and a versioned source of truth that tracks the product.

Notes

• The connector reads and writes — 24 tools, bounded by the caller's own role. Minting a key and deleting the account deliberately have no tool.
• The skills ecosystem is supported by Claude Code and claude.ai. Through other clients the connector still ships the essentials in its own tool descriptions.
• Agents can discover every documentation page at the docs' llms.txt.

Licence

MIT