lograft
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., "@lograft investigate recent error logs and correlate with git commits"
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.
lograft
Raft together logs, code, and tickets into post-mortem-ready investigation reports.
lograft is an MCP server that turns raw log query results into structured incident reports — Markdown (paste-into-ticket), JSON (machine-readable), and self-contained HTML (share offline). It composes Microsoft's official @azure/mcp for live Azure Monitor queries, and contributes the parts MS does NOT solve: KQL parsing, log↔git↔ticket correlation, PII redaction, multi-format reporting.
Works in any MCP-compatible client: opencode, Cursor, Claude Desktop, VS Code GitHub Copilot, Windsurf, Zed, Continue.dev.
Status: beta (0.1.x)
License: MIT
Runtime: Node 20.10+
Quick start
npx lograftThat's it. The MCP server starts on stdio and waits for a client to call its tools.
To use lograft as a daily tool, install globally:
npm i -g lograft@beta
lograft # starts MCP server on stdioFor live Azure mode, you also need @azure/mcp (auto-spawned via npx if absent, but installing it globally is faster):
npm i -g @azure/mcp@^2
az login # azmcp handles all Azure authRelated MCP server: elastic-mcp
What it does
Given Azure log query results (live or pasted) plus a git repo, lograft produces a single investigation bundle:
reports/<UTC-timestamp>/
├── report.md # Jira-paste-ready summary + correlations
├── data.json # machine-readable findings
└── report.html # offline-viewable, self-contained, CSP-lockedCorrelation joins are explicit-keys only: operation_Id, your configured ticket regex (e.g. [A-Z]+-\d+), and a service allowlist. Timestamp proximity is a tiebreaker, never a primary signal (no noise explosion).
Default-on PII redaction — emails, JWTs, GUIDs in auth context, Authorization headers, IPv4/IPv6, RFC1918 private ranges, internal hostnames (*.internal, *.corp, *.local). The redactor is internal middleware — there is no "skip redaction" tool surface.
MCP client setup
lograft speaks the MCP stdio transport. Below are the snippets for the four distinct config formats. Tested in opencode, Cursor, Claude Desktop; the other clients use one of these same formats — contributions welcome to confirm.
opencode (opencode.json or ~/.config/opencode/config.json)
{
"mcp": {
"servers": {
"lograft": {
"command": "npx",
"args": ["-y", "lograft@beta"]
}
}
}
}Claude Desktop family (claude_desktop_config.json / Cursor mcp.json / Windsurf / Zed)
{
"mcpServers": {
"lograft": {
"command": "npx",
"args": ["-y", "lograft@beta"]
}
}
}Paths:
Claude Desktop (macOS):
~/Library/Application Support/Claude/claude_desktop_config.jsonClaude Desktop (Windows):
%APPDATA%\Claude\claude_desktop_config.jsonCursor:
~/.cursor/mcp.json(global) or.cursor/mcp.json(project)Windsurf:
~/.codeium/windsurf/mcp_config.jsonZed: similar shape — see Zed docs
VS Code GitHub Copilot (settings.json)
{
"chat.mcp.servers": {
"lograft": {
"type": "stdio",
"command": "npx",
"args": ["-y", "lograft@beta"]
}
}
}Continue.dev (~/.continue/config.json)
{
"experimental": {
"modelContextProtocolServers": [
{
"transport": {
"type": "stdio",
"command": "npx",
"args": ["-y", "lograft@beta"]
}
}
]
}
}Tools
lograft exposes 5 fine-grained MCP tools plus 1 convenience orchestrator. Most users want lograft_investigate first. The atomic tools exist for partial pipelines.
Tool | Purpose |
| Full pipeline: parse → normalize → repo context → correlate → redact → render bundle. The tool most users want. |
| Pure: extract tables, time range, ticket mentions, projections from a KQL query. |
| Pure: turn CSV / JSON / azure-monitor-json into a unified 5-field rowset. |
| Shells out to |
| The heart: joins rows ↔ commits ↔ tickets ↔ atoms via explicit keys only. |
| Writes md+json+html bundle. Redactor middleware runs on input — non-bypassable. |
The redactor is NOT a public tool by design (plan D13). If it were optional via tool selection, a misbehaving LLM client could exfiltrate raw PII into a Jira paste. Bypassing requires explicit
redaction.bypass: trueon the render call and emits a prominent stderr warning.
Example — paste mode
You already have a Portal export CSV and want a report.
// Client calls lograft_investigate with:
{
"result": {
"kind": "inline",
"format": "csv",
"data": "timestamp,message,operation_Id,cloud_RoleName\n2026-05-22T13:58:00Z,InvalidSignature MYPROJ-42,op-1,PaymentService\n…"
},
"kql": {
"kind": "inline",
"text": "exceptions | where timestamp > ago(1h) | project timestamp, message, operation_Id"
},
"repoPath": "/path/to/your/repo",
"outDir": "./reports"
}Output:
reports/20260522-141023/
├── report.md
├── data.json
└── report.htmlreport.md opens with a ≤500-char headline summary block suitable for pasting into a ticket.
Example — live mode (delegates to azmcp)
{
"kql": {
"kind": "inline",
"text": "AppExceptions | where TimeGenerated > ago(1h) | project TimeGenerated, Message, operation_Id"
},
"live": {
"workspaceId": "<log-analytics-workspace-id>",
"subscriptionId": "<subscription-id>",
"table": "AppExceptions",
"hours": 1
},
"repoPath": "/path/to/your/repo"
}Under the hood lograft shells out to:
azmcp monitor workspace log query \
--subscription <subscription-id> \
--workspace <workspace-id> \
--table AppExceptions \
--query "AppExceptions | where ... " \
--output json \
--hours 1All Azure credentials are handled by azmcp via Microsoft's DefaultAzureCredential chain — lograft never touches AZURE_* env vars itself. See Authentication docs.
If azmcp is missing, lograft returns a clear error with install instructions.
Configuration
Place an optional lograft.config.toml in your project root (or ~/.config/lograft/lograft.config.toml). Resolution order:
--configPatharg onlograft_investigate(explicit)<MCP-process-cwd>/lograft.config.toml~/.config/lograft/lograft.config.tomlBuilt-in defaults
See examples/lograft.generic-issue-tracker.toml for a starting template covering ticket regex, service allowlist, redaction extras, and ticket-link base URL.
Architecture
┌────────────────────────┐
│ Any MCP Client (stdio) │
└──────────┬─────────────┘
│
▼
┌─────────────────────────────────┐
│ lograft (this package) │
│ parse_kql ─┐ │
│ normalize ─┼─► correlate ─┐ │
│ gather ─┘ │ │
│ redactor │ │ ← internal middleware (D13)
│ ▼ │
│ render md/json/html
└─────────────────────┬───────────┘
│ subprocess (live mode only)
▼
┌──────────────────────────────┐
│ azmcp (microsoft/mcp, GA) │
│ Owns Azure auth + KQL exec. │
└──────────────────────────────┘Trust boundaries:
lograftnever holds Azure credentials. Live mode = subprocess toazmcp.The redactor is the SOLE chokepoint between log content and any output file.
stdout is reserved for MCP JSON-RPC. All logs go to stderr. A runtime guard throws if anything else writes to stdout.
Caps and defaults
Setting | Default | Rationale |
Max correlated rows | 1000 | bounded report size |
Max commits considered | 200 | bounded git log |
Max output file size | 5 MB | bounded share-ability |
Tiebreaker window | ±10 min | timestamp proximity, AFTER key match |
| 5 min | bounded live mode |
Repo lookback | 14 days | recent context only |
All are TOML-configurable; none are removable.
Roadmap (Phase 2)
Datadog / CloudWatch / Loki / Elastic
LogSourceAdapterimplementationsSlack / Teams webhook output
Live Jira / GitHub API ticket enrichment (replaces regex-only extraction)
Public MCP registry submission to modelcontextprotocol.io
Multi-query batch mode
lograft_preview_redactiontool — read-only diff for audit
Contributing
Pull requests are welcome — especially new LogSourceAdapter implementations.
Before submitting:
pnpm install
pnpm test
pnpm typecheck
pnpm lint
pnpm buildSee CONTRIBUTING.md for adapter contracts and commit conventions.
Security reports: do not open public issues. See SECURITY.md.
License
MIT © Hoài Nhớ
Available Tools
5 toolslograft_correlateA
Join a NormalizedRowset against a RepoContext using explicit keys (operation_Id, configured ticket regex, service allowlist). Output is redacted via the internal middleware before being returned. Most users want lograft_investigate; this atomic tool is for partial pipelines.
Inputs: rowset (required), repoContext (required), joinPolicy (required), externalAtoms, +1 more.
lograft_gather_repo_contextA
Snapshot a git repository's recent commits (default: last 14 days, max 200 commits) plus current branch and origin URL. Pure read-only, shells out to git. Most users want lograft_investigate; this atomic tool is for partial pipelines.
Inputs: repoPath (required), sinceDays, maxCommits.
lograft_investigateA
Run the full investigation pipeline in one call: parse KQL (optional) -> normalize -> gather repo context -> correlate -> redact -> render md+json+html bundle. Either pass result={inline|path} (paste mode) OR live={workspace,subscription,table,...} (delegates to azmcp). Returns a Bundle with paths to the written files. This is the tool most users want first.
Inputs: repoPath (required), kql, result, live, +4 more.
lograft_normalizeA
Normalise a CSV / JSON / Azure-Monitor-JSON log export into a 5-field rowset (timestamp, level, message, source, raw). Pass sessionId to receive an opaque rowsetRef for downstream lograft_correlate calls (avoids re-shipping large payloads through MCP). Most users want lograft_investigate; this atomic tool is for partial pipelines.
Inputs: source (required), payload (required), sessionId, rowCap.
lograft_parse_kqlA
Extract structural facts from a Kusto Query Language (KQL) query without executing it. Returns the referenced tables, projected columns, time range, and ticket mentions. Most users want lograft_investigate; use this atomic tool only when you need partial pipeline output. Pure compute, no side effects, no network.
Inputs: kqlText (required), ticketRegex.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 5 tools
Each tool targets a distinct step in the investigation pipeline, with explicit guidance that most users should start with lograft_investigate. The atomic tools (parse_kql, normalize, gather_repo_context, correlate) have no overlapping purposes.
All tools share the lograft_ prefix and use clear verb_noun or verb patterns (e.g., parse_kql, gather_repo_context, investigate). The naming is uniform and predictable.
Five tools is an ideal count for this domain—neither too few nor too many. Each atomic tool serves a specific need, and the full pipeline tool ties them together efficiently.
The pipeline is fully covered: parsing KQL, normalizing logs, gathering repo context, correlating, and generating a final bundle. There are no obvious gaps for the stated purpose of log investigation.
Maintenance
Related MCP Connectors
Syslog receiver and MCP server for homelab log intelligence.
Syslog receiver and MCP server for homelab log intelligence.
MCP server for querying Forkast documentation
Repository knowledge graph MCP server for codebase understanding and debugging.
Related MCP Servers
- AlicenseCqualityDmaintenanceIncident Triage MCP is a Model Context Protocol (MCP) server for incident triage. It provides safe, auditable tools for evidence retrieval, deterministic summaries, ticket workflows, and notifications.28Apache 2.0
- AlicenseBqualityDmaintenanceSmall production-oriented MCP server for diagnosing incidents from Elasticsearch logs with unknown schema. It provides tools for log discovery, retrieval, and issue diagnosis.6MIT
- -licenseNot gradedqualityNot gradedmaintenanceMCP server that parses stack traces and logs to generate deduplicated issue drafts with severity, repro steps, and owner guesses.-
- AlicenseAqualityCmaintenanceMCP server for git archaeology — churn hotspots, blame summaries, bus factor, commit forensics and file history over any local repository.113MIT