Sparkforensics MCP

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/shuffle-works-sparkforensics-mcp

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

7 read · 1 write · 0 delete

Secrets

None asked for

Install scripts

Not scanned yet

Runs as

Node.js package

Tools (8)

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

• compare_runs

  Compare two runs: the verdict, metric deltas (metrics), categorized findings delta (findings), the stage pairs (stagePairs) with the share of executor run time they cover (runtimeCoverage), and comparability: the configuration differences, plan equality, data volume, executor capacity and the factors that bridge the wall-clock difference. confidence is ok, low or insufficient. metricDeltas and findingsDelta repeat metrics and findings under their earlier names and will be removed. The result names both Spark applications (baselineApp, candidateApp) and the build that wrote it (generator). Each…

• diagnose_run

  Description is built at runtime.

• evaluate_budgets

  Evaluate a run against pass/fail thresholds, optionally against a baseline run for regression budgets. With two runs, absolute budgets apply to the candidate (sourceB/runIdB), matching the CLI. A run with no ApplicationEnd always adds an inconclusive run-complete result, and a run with skipped log lines an inconclusive parse-incomplete result. runId is a handle for the run this server process has cached. It differs per tool by design (compare_runs returns runIdA and runIdB, evaluate_budgets the runId of its first run) and per server process, so match runs by app.id and app.attemptId, not by ru…

• get_finding_documentation

  Detection and tuning reference documentation for one finding type (not tied to a specific run), so a client without browser access can read the same background material the web view links to.

• get_finding_evidence

  Raw evidence bundle backing one finding, for drill-down after diagnose_run. Pass the runId of the diagnosis. When the runId is not cached (after a server restart), pass the source of the run, as for diagnose_run, and optionally its appId: the run is analysed again and the findingId stays the same.

• get_reference_doc

  Full tuning-reference chapter or bottleneck markdown by doc anchor (e.g. "#joins", "#bottleneck-skew", "#metric-task-duration"), so a client without browser access can read the same reference the web view shows.

• get_run_summary

  App/stage/job/sql counts, duration, and how the run ended (failed/total jobs and Spark's first-line failure reason) for a run, no findings.

• list_runs

  List candidate Spark event-log runs from a local directory or a Spark History Server, before diagnosing one with the other tools.

Signals

Stars

0

Forks

0

Open issues

0

Last commit

Oct 24, 2026 (7 days ago)

Last release

Oct 19, 2026 (11 days ago)

From GitHub (fetched Oct 24, 2026).

Details

Registry name

io.github.shuffle-works/sparkforensics-mcp

Version

0.1.0

Auth

No auth

Packages

npm sparkforensics-mcp

License

MIT

Repository

github.com/shuffle-works/sparkforensics

First seen

Oct 24, 2026

Last checked

Oct 24, 2026

Source

Official MCP Registrygithub-graphql