agentleFS
tatsoft-llctatsoft-llc/FrameworX-Projects/.claude/skills/verify-mcp-setup/SKILL.md

verify-mcp-setup

Skill1 starsChanged 26 days agoApache-2.0View on GitHub

Probe, diagnose, and fix the FrameworX MCP connection (Designer / Console / Runtime bridges) before a session depends on it. Fires on explicit user request ("verify the MCP setup", "is FrameworX connected", "the MCP tools aren't working", "check the bridges", "list solutions fails", "set up the FrameworX MCP", "can you see my Designer") AND as a pre-flight when the FIRST FrameworX tool call of a session fails with a server-not-available / connection error. Probes each bridge the session needs → identifies the failure mode from the inventory → surfaces the specific remediation → re-probes → confirms with a real bootstrap call (list_solutions / list_workspaces / runtime_connect).

  • Reads credentials

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 3 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Every capability in this workspace rides on three MCP bridges into the installed FrameworX product.

    line 21
  2. Very specific:

    When they aren't up, every tool call fails.

    line 22
  3. Distinctive:

    Only probe the arms the session needs.

    line 36
  4. Very specific:

    Call the cheapest bootstrap tool on each needed arm: list_solutions(detail='count') (Designer), list_workspaces (Console), runtime_connect …

    line 42
  5. Distinctive:

    If a first call hangs or is denied, tell the user to check for a permission dialog before diagnosing.

    line 47
  6. Very specific:

    DESIGNER_NOT_RESPONDING / pipe error mid-session: Don't re-run setup diagnosis for this one: reconnect via open_solution with the current …

    line 76
  7. Distinctive:

    Telling the user to reinstall: Every mode above has a lighter fix.

    line 99

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Verify MCP Setup: the connection pre-flight

Why this exists

Every capability in this workspace rides on three MCP bridges into the installed FrameworX product. When they aren't up, every tool call fails. And a brand-new user can't tell a missing bridge from a firewall block from a wrong host in .mcp.json. This skill mechanizes probe → diagnose → fix → re-probe so the session starts from a verified connection instead of a mystery failure. The full setup reference (same-machine, remote, VM, and the security model) is KnowledgeBase/Method/mcp-setup.md: this skill is the interactive front door to it.

The three bridges

BridgePortNeedsBootstrap probe
FrameworX AI Designer10150the FrameworX install (Designer launches on demand)list_solutions
FrameworX AI Console10151the FrameworX install (no Designer)list_workspaces
FrameworX AI Runtime10120a RUNNING TServer for the target solutionruntime_connect

Only probe the arms the session needs. A Console-only authoring session doesn't need the Runtime arm up.

Process

1. Probe

Call the cheapest bootstrap tool on each needed arm: list_solutions(detail='count') (Designer), list_workspaces (Console), runtime_connect (Runtime). A success that returns the platform context package (solutionContext / Context Package) is a PASS. The session is bootstrapped as a side effect.

Note: the very first tool call of a session may trigger a host consent prompt that is invisible to you. If a first call hangs or is denied, tell the user to check for a permission dialog before diagnosing.

2. Diagnose: failure-mode inventory

Match the error to the mode; apply the one remediation; re-probe.

  1. Connection refused / server not available (same machine). The bridge isn't running. Start it from the FrameworX install: Documents\FrameworX\Utilities\StartDesignerMCPHttp.bat (10150), StartConsoleMCPHttp.bat (10151), StartRuntimeMCPHttp.bat (10120). Each keeps a console window open: it must stay open. Start the bridge, then retry; if Claude Code was started first, the retry still works (the client reconnects per call).
  2. Bridge window reports "address already in use" when starting. A host process from an earlier session still holds the port → run the kill-zombies skill (companion repo), then start the bridge again.
  3. Timeout to a remote host. Firewall or wrong IP. On the FrameworX machine open inbound 10150/10151/10120 (only the ones used); confirm the host IP in .mcp.json matches. For a same-host VM (Parallels/VMware), keep 127.0.0.1 and port-forward instead of using the LAN IP. On FrameworX 10.1.6 or later, also confirm each bridge has an X_API_KEY set and that FX_MCP_API_KEY holds the same value on the Claude Code machine: a bridge with no key refuses every connection that is not from its own machine, VM port forwarding included (mcp-setup.md § Separate machine).
  4. .mcp.json still carries the YOUR-WINDOWS-IP placeholder (or points at localhost when FrameworX is remote). Copy .mcp.remote.example.json over .mcp.json, set the real host and, on 10.1.6 or later, the FX_MCP_API_KEY environment variable, then restart Claude Code (MCP config loads at startup).
  5. Designer/Console probes pass but the Runtime probe fails. The runtime isn't running, or the bridge's RuntimeMCPHttp.json (next to its .bat) points at the wrong TServer host/port/credentials. Start the solution's runtime first (Development profile: TServer port auto-increments from 3201: never hardcode; use the connect response). runtime_connect with explicit arguments rebinds without a bridge restart.
  6. Everything fails and the install predates 10.1.5. Pre-10.1.5 installs do not ship the MCP servers: there is no degradation path. The fix is a product update, not configuration.
  7. DESIGNER_NOT_RESPONDING / pipe error mid-session. Don't re-run setup diagnosis for this one: reconnect via open_solution with the current solution name and retry. A Designer dialog waiting on input blocks the pipe the same way, so ask the user to check the Designer window.
  8. Two clients wired to one backend (Claude Desktop .mcpb extensions AND this workspace's .mcp.json). Each client starts its own host process against the same backend, so an unclean exit leaves duplicates behind. Pick one client per backend (mcp-setup.md § One client per backend).
  9. All three servers fail to even spawn, npx / Node not found. The shipped .mcp.json launches each server via npx -y mcp-remote, so the machine running Claude Code needs Node.js (LTS) with npx on PATH. The symptom is a spawn/ENOENT error on every FrameworX server at Claude Code startup, before any port is probed. Install Node.js, restart Claude Code.

3. Confirm: the smoke test IS the bootstrap

Re-run the step-1 probes until the needed arms PASS. State the result concretely: "Designer bridge up. list_solutions returned N solutions and the platform context package." Do not declare setup verified from a ping or a port check alone: only a successful tool round-trip counts. Never build a mock or proceed "offline" as a workaround; connect for real first.

Anti-patterns

  • Diagnosing before probing. One cheap call tells you which arm is broken; start there.
  • Widening timeouts or retry loops on a refused connection. Refused = not running; fix the cause.
  • Probing all three arms when the session needs one. Cost discipline: probe what the work requires.
  • Telling the user to reinstall. Every mode above has a lighter fix; reinstall is never step one.
  • Skipping the consent-prompt warning on the first call of a session.

Composition

  • kill-zombies: mode 2 (held port) and any .dbsln lock found on the way route there. Ships in the companion repo.
  • The browser side: Claude Code's built-in browser support handles the browser connection; the web-client test recipe is KnowledgeBase/Patterns/Test-Web-Displays-in-Chrome.md.
  • new-project-bootstrap / ground-first: both assume a bootstrapped session; this skill is what runs when that assumption fails.
  • KnowledgeBase/Method/mcp-setup.md, the full reference this skill fronts, including the auth model (plain HTTP, with an API key for network clients from 10.1.6; the security boundary is the network; localhost or trusted LAN/VPN only).

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.