Breakdown includes an installable Agent Skill and a local MCP server so tools like Codex and Claude can recognize when connectivity may explain a failure, then ask for current network health, recent issues, compact app/service evidence, and structured Breakdown analysis.
Connectivity errors are symptoms, not diagnoses. ECONNRESET, timeouts, failed DNS lookups, and interrupted TLS connections can come from the local network, the Internet path, a proxy or VPN, runtime configuration, security software, or the upstream service.
An agent should query Breakdown when timing or reachability evidence could separate those causes, especially after a transient failure has passed. Breakdown can often establish whether the failure was on the local network, Internet path, or app side. The agent can then compare runtime, configuration, authentication, and other evidence that Breakdown cannot observe directly.
Record the exact error and timestamp.
Ask Breakdown for current health and nearby issues.
Compare Wi-Fi/LAN, Internet-path, and app/service evidence.
Check runtime, proxy/VPN, DNS/TLS, authentication, and upstream service evidence that Breakdown cannot observe directly.
Recommend the smallest reversible next step and preserve the evidence.
Agent Skill
Teach your agent when to reach for Breakdown
The open Breakdown connectivity skill works across Agent Skills-compatible tools. It helps an agent recognize connectivity-related failures, use Breakdown when it is already connected, and set up the app and MCP bridge when needed.
Choose one installation path.
Universal skill installFor Codex, Claude Code, and other supported agents.
The app exposes MCP tools for network health, recent issue summaries, evidence selectors, and cloud-backed Breakdown analysis. The goal is compact structured output first, with detailed evidence available when the agent needs to drill in.
Current health, segment scores, freshness, and issue pointers
Recent network issue summaries with deterministic metadata
Shared trace, app-service, and evidence selector paths
Async Breakdown analysis runs that can be started and polled
Failure context
See now and the incident window
An agent can start with current health and recent issues, then request bounded app, endpoint, or route evidence around the relevant time.
Local by design
Network evidence only
The server listens on your Mac's loopback interface, requires the current user's generated authorization, and exposes bounded Breakdown evidence and analysis tools. It cannot read arbitrary files, logs, settings, account data, or Keychain contents.
MCP connection
Connect your agent to the running app
Keep Breakdown running. For Claude-style clients, choose Copy MCP config for AI tools from the gear menu or Settings, paste the JSON into the client, and reload it. Codex uses its own setup command, shown below. The local server is enabled by default and can be disabled in Settings.
01
Run Breakdown
The stdio bridge connects to the running app. Breakdown remains the single owner of monitoring and evidence.
02
Add the bridge
Use Breakdown's copied JSON in Claude-style clients. For Codex, add the installed bridge with the command below.
03
Reload your client
Reload the MCP client, then confirm that it discovers Breakdown's tools.
Codex CLIUse this command for the standard app installation.
The API timed out. Check the path before blaming the API.
An agent can ask Breakdown for current health and recent issues, inspect app or route evidence, and start a structured Breakdown analysis when the failure needs a deeper answer.
Agent The upload failed with ECONNRESET. Check Breakdown before retrying.
Breakdown LAN is healthy. Internet path loss increased during the failed upload. The app endpoint remained reachable.
Agent Hold the retry and gather the relevant path evidence.
