mcp-pairinglisted
Install: claude install-skill sananthanarayan/skilldrop
# mcp-pairing
Turns a loop, workflow or task into a per-stage connection plan: which stages get an MCP
server, which get none, and the exact access each one holds. An MCP server is a capability
handed to the agent with someone's credentials, so the default answer for every stage is
**fewer servers, read-only, scoped to one project**, and every grant beyond that is argued
from the stage's output. The plan stays honest about names: it never invents a server,
package or URL.
Complements `agent-threat-model`, which audits a finished capability surface for data
exfiltration paths. This skill decides what the surface should be in the first place;
that one attacks it.
## How to respond
**Four rules come before the steps and outrank them:**
- **Answer what was asked, first.** Open with the plan in plain words: how many servers, which ones, read or write, and which stages need none. The table comes after.
- **Use only what you were given.** Name only the tools, servers, stages and scopes the input contains, plus vendors' own official servers under the naming rule in step 3. Never invent a package, endpoint, URL or version. What the plan needs and doesn't have is listed once at the end under "To confirm".
- **Deliver from what you have.** When the request names a loop, workflow or task, state the assumptions in a line and produce the plan. When it names nothing to pair, ask for it in one or two plain sentences and say what the plan will contain once it arrives.
- **Write for someo