Backend Investigator

Install

Works with Claude Code, Codex, VS Code, Cursor, Windsurf, Claude Desktop. Read the repository before you give any tool access to your data.

npx agentmarket install connector/mrzedov-backend-investigator

About

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this server can do and what it asks for, from its tool list and install config. Checked Oct 25, 2026.

Tools

11 read · 0 write · 0 delete · 3 not yet classified

Secrets

None asked for

Install scripts

Not scanned yet

Runs as

Node.js package

Tools (14)

Read from the source code; the running server may differ. Descriptions are the server’s own words; hidden characters are shown as [U+…].

• get_investigation

  Retrieve an immutable investigation snapshot by investigationId, with paginated summarized evidence, findings, accepted hypotheses and source availability. Evidence references may point to another page. No live source or deleted project credentials are required.

• get_latest_deployment

  Read latest/previous GitHub deployments using success-status timestamps, or explicitly enabled local demo-only simulated switches. Releases are publication markers, never claimed rollout times. Build historical before/after windows without changing infrastructure.

• get_project_capabilities

  Discover enabled sources and available read-only checks for a project before choosing follow-up tools. Source IDs are included, connection configuration is not.

• inspect_database

  Read PostgreSQL health, current sessions and blocking locks. For a performance review explicitly request performanceChecks: query_statistics (pg_stat_statements), tables, indexes, vacuum, io. Optional checks use bounded fixed read-only queries and report availability/reasons individually. performanceLimit defaults to 5 rows per sample (maximum 20); cumulative statistics are not a recent time window. No arbitrary SQL, EXPLAIN ANALYZE, query cancellation or maintenance is exposed.

• inspect_metrics

  Read fixed, scoped Prometheus range queries: availability (scrape targets), http (request rate, 5xx percent, classic histogram p95/p99) and resources (process CPU cores and resident bytes). No arbitrary PromQL or writes. Windows are at most 24 hours with at most 21 evaluation points. Missing data/non-finite values are explicit. Optional equal-length earlier comparison windows report sampled-mean deltas, not causal proof or whole-window percentiles.

• inspect_queue

  Read BullMQ counts, worker visibility and sanitized bounded job samples for a configured queue. No job retry, removal or state changes are exposed.

• inspect_redis

  Read Redis health and sanitized SLOWLOG samples for a configured source. Never scans all keys, reads values or executes write commands.

• inspect_service

  Follow up on a Docker service/container by its exact serviceId and sourceId. Reads runtime inventory, health, optional stats and recent restarts; never restarts or modifies containers.

• investigate_backend

  Primary high-level tool: conservatively select a strategy from an unambiguous question or explicit strategy. Unknown/mixed questions preserve full health checks. Conditional checks require observed dependency hints; irrelevant sources are skipped. Returns execution decisions, facts and optional reasoning; configured history appends a local immutable archive.

• investigate_deployment_regression

  Compare evidence before/after an observed successful deployment or explicitly labelled release marker. Uses compatible archived snapshots and historical logs; missing baselines are not zero. Separates changes, unchanged metrics, time correlation and optional hypotheses. Current follow-up checks are not historical evidence. Appends an immutable local archive when configured.

• investigate_operation

  Investigate a specific endpoint or operation in one service/log source and explicit time window. Match route or query, then correlate sampled request/trace IDs. Live infrastructure anomalies are only temporally related, not proof of causation. When history is configured, appends a local archive without changing connected infrastructure.

• list_investigations

  List immutable saved investigation summaries with pagination and optional projectId filter. Historical projects need not still exist; credentials and raw logs are not loaded.

• list_projects

  List configured projects with pagination. Use this first to select a projectId; no connection credentials are returned.

• search_logs

  Search bounded Docker or scoped Loki logs by optional serviceId and literal substring query. For Loki, serviceId is the source ID, not a container ID; configured labels cannot be overridden. Use a narrower explicit time window/query if sampled or oversized.

Signals

Weekly downloads

51

Last release

Oct 13, 2026 (16 days ago)

Last package publish

Oct 13, 2026 (16 days ago)

From the package registry.

Details

Registry name

io.github.mrZedov/backend-investigator

Version

1.7.1

Auth

No auth

Packages

npm @mr.zedov/backend-investigator

License

License unknown

First seen

Oct 24, 2026

Last checked

Oct 24, 2026

Source

Official MCP Registry