agentleFS
MicrosoftKnown publishermicrosoft/testfx/.agents/skills/troubleshoot-binlog-mcp/SKILL.md

troubleshoot-binlog-mcp

Skill1.1k starsChanged 23 days agoMITView on GitHub

Diagnose dotnet-msbuild binlog MCP startup failures. USE FOR: CONNECTION_CLOSED, binlog server won't connect, missing binlog tools, Microsoft.AITools.BinlogMcp not found, dnx launch errors. Checks workspace SDK, feeds, and source mapping without bypassing policy. DO NOT USE FOR: errors inside a binlog (use binlog-failure-analysis), capturing binlogs (use binlog-generation), unrelated MCP servers.

We split this skill into the rules it gives an agent

20 rules. About 0 are common, 12 are distinctive, and 8 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈12

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 8

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    No binlog file or project build is needed.

    line 11
  2. Very specific:

    Use the user's error and any host MCP status or startup logs already available.

    line 18
  3. Distinctive:

    If the binlog server is connected and its tools are discoverable, stop startup troubleshooting.

    line 22
  4. Distinctive:

    If the plugin's skills loaded but its binlog server did not, continue below.

    line 24
  5. Distinctive:

    If the plugin itself is not installed/enabled or no server is registered, address that host configuration first.

    line 25
  6. Distinctive:

    If execution or host logs are unavailable, explain that limit and give the next command for the user to run.

    line 27
  7. Distinctive:

    Run diagnostics from that workspace with the host's environment, not from the plugin directory or a temporary directory with different …

    line 39
  8. Very specific:

    Check the applicable global.json and inherited NuGet configuration, including parent/user configuration, clear, disabled sources, and …

    line 48
  9. Distinctive:

    Do not dump credential settings, private feed URLs, tokens, or credential-provider output into the answer.

    line 54
  10. Distinctive:

    Prefer an existing startup log from the failing host.

    line 59
  11. Distinctive:

    Only if package availability remains unclear, run this supporting probe under the same workspace settings with a bounded execution:

    line 71
  12. Distinctive:

    Do not repeat expensive probes once the evidence establishes the next action.

    line 91
  13. Very specific:

    Add nuget.org only if organizational policy permits and the user approves the change.

    line 101
  14. Very specific:

    Do not remove clear, disable source mapping, or override sources to bypass enterprise policy.

    line 102
  15. Very specific:

    If appropriate and approved, install the tool from an approved feed and register its binlog-mcp executable with the host as an alternative …

    line 103
  16. Distinctive:

    Do not edit NuGet settings, host registration, or SDK pins without the user's approval.

    line 109
  17. Very specific:

    If policy prevents recovery, say so.

    line 110
  18. Very specific:

    After an approved change, reconnect/restart the server using the host's supported controls.

    line 115
  19. Very specific:

    Use connected only after the host discovers the server's tools, failed for an observed launch/connection failure, and unverified when only …

    line 126
  20. Distinctive:

    Never substitute "package found" or "tool installed" for a verified MCP connection.

    line 128

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 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

Troubleshoot Binlog MCP Startup

Explain why the plugin's binlog MCP server cannot connect and give an evidence-based, policy-safe recovery. This diagnoses the server launch, not failures recorded in a build. No binlog file or project build is needed.

This skill runs on demand. It cannot see a host error that was never exposed to the agent, inject session-start context, or guarantee activation merely because a plugin was installed.

1. Establish what failed

Use the user's error and any host MCP status or startup logs already available. CONNECTION_CLOSED alone says that the transport closed, not why. Do not translate it directly into "package missing."

  • If the binlog server is connected and its tools are discoverable, stop startup troubleshooting. Continue the original binlog analysis; do not change a working setup.
  • If the plugin's skills loaded but its binlog server did not, continue below.
  • If the plugin itself is not installed/enabled or no server is registered, address that host configuration first. Do not diagnose NuGet package resolution without a launch attempt.
  • If execution or host logs are unavailable, explain that limit and give the next command for the user to run. Never claim to have inspected logs or repaired a connection.

2. Match the host's workspace and SDK

Discover the affected workspace directory and the host's actual MCP command, arguments, and environment from available configuration. The shipped command is:

dotnet dnx Microsoft.AITools.BinlogMcp --yes --prerelease

Run diagnostics from that workspace with the host's environment, not from the plugin directory or a temporary directory with different NuGet settings. Inspect:

dotnet --info
dotnet --version
dotnet nuget list source

Check the applicable global.json and inherited NuGet configuration, including parent/user configuration, clear, disabled sources, and packageSourceMapping. A workspace clear can replace the default nuget.org source with a private mirror that lacks this package. An installed SDK is not necessarily the SDK selected in this directory. The shipped dotnet dnx command requires a .NET 10+ SDK.

Do not dump credential settings, private feed URLs, tokens, or credential-provider output into the answer. Inspect locally as needed and report a redacted source label and relevant error.

3. Capture the real launch error

Prefer an existing startup log from the failing host. Otherwise reproduce its unchanged launch command once, from the affected workspace, capturing both stdout and stderr plus the exit code. Do not initially add --source, replace the configuration, remove global.json, clear caches, or install another copy: those change the conditions being diagnosed.

A stdio MCP server may remain running waiting for protocol input. Keep stdin open while observing startup and use a bounded execution (for example, 30 seconds) with cleanup of only the diagnostic process tree, including credential-provider children. Use the host's supported timeout/cancellation facility; if it cannot safely clean up, ask the user to run the command locally instead. Do not leave a background server running or kill unrelated dotnet processes. A timeout, silence, or an open process is not proof that MCP initialized.

Only if package availability remains unclear, run this supporting probe under the same workspace settings with a bounded execution:

dotnet package search Microsoft.AITools.BinlogMcp --exact-match --prerelease --format json

Inspect the exact package ID and version in searchResult[].packages, and check both top-level and per-source problems. Exit code zero alone does not establish a successful search.

Observed evidenceDiagnosis and next action
dnx reports no matching version in the configured feedsThose settings did not resolve the package. Check whether the approved mirror provides prerelease versions of this exact package.
Search returns no packages and no source problemsOnly the searched feeds lack a listed match. This is not proof that dnx must fail: a cached tool or unlisted version may still work.
Search has top-level/per-source problems, HTTP 401/403, TLS, network, or credential-provider errorsAvailability is unverified, not "package missing." Address the specific feed/access failure first, even if search exited zero.
Search finds the package but dnx failsFollow the launch error. Search does not apply package source mapping; check mapping, restore/download access, and runtime requirements. Do not declare the connection healthy.
Missing dotnet, unrecognized dnx, or SDK selection errorFix the host's PATH/SDK selection. Account for global.json; do not silently upgrade the repository or remove its SDK pin.
Timeout, unsupported search command, or malformed/non-JSON outputReport an inconclusive probe. Use launch logs or a user-run diagnostic; do not infer absence or success.
Launch succeeds but host connection still failsInvestigate the host's actual command/environment and MCP initialization logs; package discovery is no longer the demonstrated failure.

Do not repeat expensive probes once the evidence establishes the next action.

4. Choose a policy-safe recovery

Recommend the smallest change supported by the observed failure:

  • Have the feed owner mirror Microsoft.AITools.BinlogMcp and its required versions/dependencies into an approved feed, or correct an approved mapping that excludes it.
  • Repair the approved feed's authentication, connectivity, or certificate configuration when that is the observed problem. Never suggest disabling certificate validation.
  • Add nuget.org only if organizational policy permits and the user approves the change. Do not remove clear, disable source mapping, or override sources to bypass enterprise policy.
  • If appropriate and approved, install the tool from an approved feed and register its binlog-mcp executable with the host as an alternative MCP server. Confirm the executable's path and avoid registering a duplicate server. Installation alone is not connection recovery.
  • For SDK/PATH issues, make the required SDK available to the host or use a deliberately configured alternative launch. Ask before changing repository SDK selection.

Do not edit NuGet settings, host registration, or SDK pins without the user's approval. If policy prevents recovery, say so. For an existing binlog, offer the text-log replay fallback in binlog-failure-analysis rather than blocking build diagnosis on MCP repair.

5. Verify and report

After an approved change, reconnect/restart the server using the host's supported controls. Confirm MCP initialization and a tool listing that contains binlog tools such as binlog_overview. If a binlog is available, resume the original analysis.

Report the observed cause (or what remains unknown), the relevant redacted evidence, the smallest recovery, and what was actually verified. End with one status line:

MCP status: connected

Use connected only after the host discovers the server's tools, failed for an observed launch/connection failure, and unverified when only probes or advice are available. Never substitute "package found" or "tool installed" for a verified MCP connection.

Download now (ZIP)

Trending

Quiet since September.about the whole repository

Dec 2025: 1 mentionJan 2026: 0 mentionsFeb 2026: 0 mentionsMar 2026: 14 mentionsApr 2026: 0 mentionsMay 2026: 3 mentionsJun 2026: 0 mentionsJul 2026: 0 mentionsAug 2026: 0 mentionsSep 2026: 2 mentionsOct 2026: 0 mentionsNov 2026: 0 mentions
Dec 2025Jun 2026Nov 2026
Mentions a month of the microsoft/testfx repository that our weekly crawl found. There is no adoption data for this file yet.

Who's talking about microsoft/testfx

The biggest 5 of 68 mentions linking to the repository that our crawl found. Each line is their own headline.

See all 68

Talked about on GitHub and Stack Overflow (68 times).

On GitHub · 41

36 more on GitHub

On Stack Overflow · 27

22 more on Stack Overflow

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.