crash-curator-mcp
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@crash-curator-mcpWhat caused the most recent crash?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
crash-curator-mcp
A small MCP server that turns systemd-coredump and journal evidence into bounded, normalized crash reports before an AI model sees it.
It is intentionally not a general-purpose log reader. Every query has fixed time, row, line, and byte budgets. The server uses Bun 1.4 APIs directly (Bun.spawn, Bun.file, and Bun streams), has no runtime dependencies, and does not use Node.js APIs.
Requirements
Linux with systemd 257 or newer
Permission to read the user's coredump and journal metadata
The npm package contains a standalone Linux x64 executable, so Bun is not required at runtime. Building from source requires Bun 1.4 or newer.
Related MCP server: Crash MCP Server
Tools
crash_list: scans at most 50 coredump records and returns at most 10 normalized entries.inspect_crash: returns one crash, stack highlights, matching-crash frequency, warnings and process logs from a 30-second window, OOM evidence, and package changes from the preceding 24 hours.
inspect_crash has a hard 12 KB response budget. Common tokens, passwords, API keys, and URL credentials are redacted. It never extracts a core or invokes a debugger.
Install
bun add --global crash-curator-mcpThe package published to npm contains only the compiled executable plus npm's required package metadata, README, and license. Source code and tests are excluded from the tarball.
Develop
bun test
bun run build
./dist/crash-curator-mcpThe server speaks MCP over stdio. Running it directly waits for JSON-RPC input; normally an MCP client starts it.
Codex
Add this to the user-level ~/.codex/config.toml:
[mcp_servers.crash-curator]
command = "crash-curator-mcp"Claude Code
claude mcp add --scope user crash-curator -- crash-curator-mcpOpenCode
Add a user-level entry under mcp in ~/.config/opencode/opencode.json:
{
"mcp": {
"crash-curator": {
"type": "local",
"command": ["crash-curator-mcp"],
"enabled": true
}
}
}MCP has no universal cross-client registry. “Global” means configuring the server once at user scope in each client, rather than embedding it in a project or in Omarchy.
Privacy and scope
Commands are executed without a shell.
Journal queries are limited to 30 seconds around the crash.
Package history is limited to the 24 hours preceding the crash.
Core memory is never read or copied.
Raw logs are never returned.
The output reports when evidence was omitted or truncated.
Symbolization should be a separate, explicit escalation because a core can contain credentials and private documents and debugger output can be very large.
This server cannot be deployed
Maintenance
Related MCP Connectors
Tenant-scoped evidence intake and controlled AI context over MCP.
Remote MCP for C2PA intake verifier MCP, structured receipts, audit logs, and reviewer-ready evidenc
Evidence-readiness MCP server: validate, audit, and score briefs, memos, and evidence packs.
Remote MCP for A2A failure replay MCP, structured receipts, audit logs, and reviewer-ready evidence.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceEnables MCP-capable applications to access, manage, and analyze system core dumps through integration with systemd-coredump functionality.20 npm3MIT
- AlicenseBqualityDmaintenanceEnables AI assistants to analyze Linux system crash dumps by automatically discovering dump files, matching kernels, and executing interactive crash analysis commands through the MCP protocol.67MIT
- AlicenseNot gradedqualityBmaintenanceA read-only MCP server that exposes Kubernetes cluster telemetry tools (pods, events, logs, metrics, ArgoCD syncs) with automatic redaction for incident triage and hypothesis ranking.MIT
- AlicenseAqualityBmaintenanceEnables MCP-capable clients to query coding-agent conversation exports and retrieve token-bounded evidence bundles with source provenance and hash verification, so LLMs can ground responses in verifiable history without replaying full transcripts.5MIT