convez.ai
MCP ServerDados & AnalyticsIntermediário
arikanatakan

DataXPlan MCP

arikanatakan/dataxplan-mcp

Analisa planos PostgreSQL EXPLAIN para identificar gargalos e sugerir otimizações

Ver no GitHub
InicialPluginScore · de 0 a 100

Servidor MCP que analisa planos EXPLAIN do PostgreSQL, identificando gargalos, erros de estimativa e sugerindo otimizações. Fornece análise estruturada, comparação de planos e gráficos de auto-tempo para agentes de IA, sem nunca conectar ao banco de dados.

Instalação

$npx convez-ai arikanatakan/dataxplan-mcp

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
2
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Diagnóstico automatizado de queries lentas em PostgreSQL
  • Comparação de planos de execução para detectar regressões
  • Geração de recomendações de índices baseadas em análise estruturada
  • Integração com agentes de IA para otimização de banco de dados

Veredito

Ferramenta sólida e bem arquitetada para análise de performance em PostgreSQL. Preenche uma lacuna real ao automatizar a leitura de planos EXPLAIN, que frequentemente confunde agentes de IA, delegando a interpretação complexa a uma biblioteca versionada e determinística.

README

A tradução para o português deste README ainda está na fila.

dataxplan-mcp

CI PyPI License: MIT

An MCP server that exposes dataxplan, the PostgreSQL EXPLAIN-plan analyzer for Python, as tools for AI agents: give it the output of EXPLAIN (ANALYZE, BUFFERS, FORMAT JSON) and it returns the bottlenecks, the estimation errors, documented findings with a suggested action and a source reference, a regression comparison, and a self-time chart.

Agents asked why a query is slow tend to eyeball the plan and get it wrong: self time is per loop and inclusive of children, so the slow node is rarely the obvious one, and a row mis-estimate (the usual root cause) is buried in the output. Reading the plan belongs in a deterministic, versioned library that the agent calls, which leaves the agent to interpret the result and decide. The server never connects to a database: the agent runs EXPLAIN and passes the output, so nothing leaves its environment.

dataxplan-mcp architecture: an AI agent calls the server's analysis and chart tools, which route to the validated dataxplan core and return structured JSON or a PNG chart

Tools

Analysis tools return dataxplan's payload: the metrics, the findings (each with a severity, a suggestion and a documented source reference) and a summary.

ToolPurpose
analyze_planbottlenecks, estimation errors and findings from an EXPLAIN plan (JSON, text, YAML or XML)
compare_planscompare two plans for regression (timing, shape, estimates, findings)
plan_treethe plan as an annotated text tree (self time, rows, flags per node)
describe_inputshow to produce the plan, the accepted formats, the findings and the thresholds

Chart tools return a PNG image.

ToolPurpose
plan_chartself time per node, with the high-severity findings highlighted

All tools are read-only, and the server makes no database connection.

Installation

Run it with uv (no install needed):

uvx dataxplan-mcp

or install from PyPI:

pip install dataxplan-mcp

Configuration

Add it to your MCP client. For example:

{
  "mcpServers": {
    "dataxplan": {
      "command": "uvx",
      "args": ["dataxplan-mcp"]
    }
  }
}

If you installed with pip, use "command": "dataxplan-mcp" with no args.

Example

analyze_plan(plan="<EXPLAIN (ANALYZE, BUFFERS, FORMAT JSON) output>")
  -> { "summary_metrics": { "execution_time_ms": 1240.0,
                            "max_estimation_error": 20000, ... },
       "findings": [ { "id": "seq_scan_hot", "severity": "high",
                       "detail": "... 95% of execution time ...",
                       "suggestion": "consider an index ...",
                       "reference": "PostgreSQL: Using EXPLAIN; the Indexes chapter" } ],
       "summary": "dataxplan - ...\n  execution time ..." }

The agent runs the EXPLAIN itself and pastes the output (any format) as plan.

Design

The server is a thin, stateless wrapper. All of the analysis lives in the dataxplan library, which computes the metrics from the documented EXPLAIN fields and grounds each finding in the PostgreSQL manual (and Leis et al. 2015 for the estimation rules). The server adds the tool schema, read-only annotations and an input-schema helper so an agent can format the input and act on the result. The findings are documented heuristics, not guarantees, and the server connects to nothing.

Related

License

MIT. Written and maintained by Atakan Arikan, MSc Student at Tsinghua University and Politecnico di Milano.

Ver o original no GitHub

Bom para

Relacionados