byaml-mcp
# byaml-mcp v2 (ByaML-004)
MCP server de BYaML Architecture Graph para agentes (Kiro, Cursor, Claude, Copilot).
**El agente NO escribe YAML — usa el grafo.** Input: evaluación/scan/terraform plan.
Output: **Architecture Graph v0.4** + insights (blast radius, costo) + remediación.
## Tools (8)
| Tool | Descripción |
|---|---|
| `graph_from_evaluation` | eval_id/findings → Graph v0.4 |
| `graph_from_terraform` | tfplan.json → Graph v0.4 |
| `graph_from_account` | profile AWS → Graph v0.4 (scan) |
| `validate_graph` | valida grafo (byaml.validate) |
| `blast_radius` | downstream impact de un nodo (nan-graph) |
| `cost_summary` | costo por layer/servicio/owner (nan-graph) |
| `remediation_plan` | findings → pasos IaC (snippet terraform) |
| `schema_latest` | fetch version/catalog del registry schema.byaml.org (ByaML-003) |
## Run
```bash
uv run server.py # stdio
uvx byaml-mcp # publicado (PyPI)
```
## Config MCP (Kiro / Claude / Cursor)
```json
{ "mcpServers": { "byaml": { "command": "uvx", "args": ["byaml-mcp"] } } }
```
## Depende de
- `byaml` (PyPI 0.4.0a1) — modelo v0.4 + convert (from_evaluation/from_terraform) + validate
- `nan-graph` (PyPI) — blast_radius, cost_chain, team_cost, SPOF
- `mcp` (FastMCP) — transporte stdioTDQS
Scored across 8 tools
Each tool targets a distinct concern: schema retrieval, graph construction from three different input sources, graph validation, downstream impact analysis, cost aggregation, and remediation planning. No two tools overlap in purpose, making selection unambiguous.
Names follow snake_case and mostly verb_noun (validate_graph, cost_summary, remediation_plan), with graph_from_* indicating source. Minor deviations like schema_latest and blast_radius are still clear and consistent in style, so no confusion arises.
With 8 tools, the surface is well-scoped for an architecture analysis server. Each tool contributes to a clear workflow (create graph, validate, analyze, remediate), and the count is within the ideal range for coherence.
The set covers graph generation from multiple sources, validation, blast radius, cost, and remediation. Minor gaps exist (e.g., no graph diff or editing), but the core lifecycle for analysis is complete and no dead ends are apparent.