log-ai-compressor

{
  "_meta": {
    "io.modelcontextprotocol.registry/official": {
      "status": "active",
      "isLatest": true,
      "updatedAt": "2026-10-24T03:44:15.666279Z",
      "publishedAt": "2026-10-24T03:44:15.666279Z",
      "statusChangedAt": "2026-10-24T03:44:15.666279Z"
    }
  },
  "server": {
    "name": "io.github.hu-chenyu/log-ai-compressor",
    "_meta": {
      "io.modelcontextprotocol.registry/publisher-provided": {
        "tool": "mcp-publisher",
        "tools": [
          {
            "name": "analyze_log_file",
            "description": "Analyze one or more local log files (merged oldest-first by mtime, suitable for rotated logs). Returns error clusters ranked by priority, each with a root-cause confidence tier (CONFIRMED / LIKELY / INSUFFICIENT) and an explicit list of what evidence is still missing. Runs fully on-device."
          },
          {
            "name": "analyze_log_text",
            "description": "Analyze a log text directly — pasted content, command output, or a fragment an agent already read. Same clustering and root-cause confidence output, without touching the filesystem."
          },
          {
            "name": "compare_log_files",
            "description": "Compare 2-5 log files and list errors that are new, disappeared, or common relative to a baseline file. Answers 'what changed between this deploy and the last one'."
          },
          {
            "name": "export_report",
            "description": "Export analysis as a compressed report in md / json / text / html / summary format. Markdown and summary are tuned for pasting into an LLM: a few hundred tokens instead of millions of raw log lines, with every finding still carrying its line numbers for citation."
          },
          {
            "name": "get_cluster_detail",
            "description": "Retrieve the full evidence for one error cluster: representative raw log line, surrounding context lines, de-noised stack trace, instance line-number index, and the reasoning behind its root-cause verdict."
          },
          {
            "name": "list_rules",
            "description": "List built-in log parsing rule templates and the log formats each targets. Check this first when a log parses badly."
          },
          {
            "name": "verify_evidence",
            "description": "Re-read the source log and re-verify every evidence reference in a report: file, line range, sha256 digest. Returns one of four verdicts - VERIFIED (every reference matches), TAMPERED (content really changed), UNVERIFIABLE (cannot check right now, e.g. the log is momentarily locked by another process - NOT an accusation of tampering), NO_EVIDENCE. This is the tool's trust anchor - the agent does not need to believe the analysis, it can re-run this step itself."
          },
          {
            "name": "check_environment",
            "description": "Self-check with no arguments: confirms the tool is importable and reports version, available built-in rules, and whether optional AI interpretation is configured."
          }
        ],
        "privacy": {
          "note": "All analysis runs on-device. The server only reads local file paths the calling agent passes in; no network calls are made unless the optional AI-interpretation layer is explicitly configured.",
          "dataEgress": "none"
        },
        "benchmark": {
          "note": "Measured on 10 Loghub-2.0 datasets (ISSTA'24) whose event templates are human-annotated. At an equal 32k-token budget, evidence retention vs Drain3 is 7 wins / 1 tie / 2 losses - the two losses (HPC, Mac) are documented with root causes rather than hidden. Output size stays bounded in the low hundreds of tokens regardless of log size (re-measured 313-788 tokens on 120k-180k line logs), whereas Drain3's output grows from 317 to 566,802 tokens. Full methodology and limitations: https://github.com/hu-chenyu/log-ai-compressor/blob/main/docs/BENCHMARK.md",
          "source": "https://zenodo.org/record/8275861"
        },
        "toolVersion": "1.8.1"
      }
    },
    "title": "log-ai-compressor",
    "$schema": "https://static.modelcontextprotocol.io/schemas/2025-12-11/server.schema.json",
    "version": "2.12.9",
    "packages": [
      {
        "version": "2.12.9",
        "transport": {
          "type": "stdio"
        },
        "identifier": "log-ai-compressor",
        "runtimeHint": "uvx",
        "registryType": "pypi",
        "registryBaseUrl": "https://pypi.org",
        "runtimeArguments": [
          {
            "name": "--from",
            "type": "named",
            "value": "log-ai-compressor[mcp]"
          },
          {
            "type": "positional",
            "value": "log-ai-compressor-mcp",
            "valueHint": "command"
          }
        ]
      }
    ],
    "repository": {
      "url": "https://github.com/hu-chenyu/log-ai-compressor",
      "source": "github"
    },
    "websiteUrl": "https://github.com/hu-chenyu/log-ai-compressor",
    "description": "Local log forensics: compress logs into citable evidence with root-cause confidence tiers."
  }
}