Iron Rod

iron-rod

active

Repo path (home): /home/eric/projects/misc/iron-rod
Repo path (work): —
Completion: 90%
Estimated hours remaining: 6.0
Confidence: medium
Last analyzed: 6/23/26, 3:45 AM
Target state: API-first local RAG over LDS scriptures and General Conference with tier-aware retrieval and anti-hallucination guardrails, plus a thin CLI/MCP wrapper — accurate, guardrailed, tested.

Automation

Enable automation targetInterval (days)

CLAUDE.md auditInterval (days)

Docker auditInterval (days)

Makefile auditInterval (days)

Pipeline auditInterval (days)

Tasks

Implement thin MCP server wrapping the /v1/answer API

Create MCP server module with answer + search tool definitions

Wire tools to call the API/core answer pipeline and map Answer -> MCP response

Add entry point / console script and packaging metadata

Unit tests for tool registration, param passthrough, and refusal/citation handling

Run history

Started	Trigger	Status	Duration
6/23/26, 3:44 AM	cron	success	1m 3s
6/22/26, 4:24 AM	cron	success	1m 36s

View all →

Last raw analysis JSON

JSON

{
  "confidence": "medium",
  "operational": {
    "kind": "none",
    "drift": "n/a",
    "score": null,
    "checks": [],
    "status": "not_applicable",
    "latest_tag": "v0.1.0",
    "deployed_version": null
  },
  "gap_analysis": "The definition of done calls for a thin CLI/MCP wrapper. The CLI exists, but no MCP server is implemented anywhere in the repo — packages/api/__init__.py references an MCP wrapper that calls the endpoints, but the actual server, tool definitions, and tests are absent. This is the only verifiable, code-level gap remaining; the RAG core, tier-aware retrieval, anti-hallucination guardrails, API, and test suite are all in place.",
  "proposed_tasks": [
    {
      "title": "Implement thin MCP server wrapping the /v1/answer API",
      "subtasks": [
        {
          "title": "Create MCP server module with answer + search tool definitions",
          "estimated_minutes": 150
        },
        {
          "title": "Wire tools to call the API/core answer pipeline and map Answer -> MCP response",
          "estimated_minutes": 90
        },
        {
          "title": "Add entry point / console script and packaging metadata",
          "estimated_minutes": 30
        },
        {
          "title": "Unit tests for tool registration, param passthrough, and refusal/citation handling",
          "estimated_minutes": 90
        }
      ],
      "description": "Add the missing MCP wrapper promised by the definition of done and referenced in packages/api/__init__.py. Create an MCP server (e.g. packages/mcp or a module under api) exposing the RAG capabilities as MCP tools — at minimum an `answer` tool and a `search` tool — that call the existing FastAPI /v1/answer (and retrieval) surface rather than re-implementing the pipeline. Keep it thin: forward query/tier/source-class/top_k params, surface the structured Answer (text, citations, refusal status) back through MCP. Register an entry point/console script and document usage. Include unit tests mocking the HTTP/core layer to verify tool registration, parameter passthrough, and refusal/citation passthrough.",
      "estimated_minutes": 360
    }
  ],
  "completion_percent": 90,
  "current_state_summary": "Iron Rod has a complete API-first RAG core (ingest, hybrid retrieval, tier-aware scoring, optional cross-encoder rerank, citation/grounding/hallucination guardrails, eval harness) in packages/core, a FastAPI service exposing /v1/answer plus health/stats/version in packages/api, and a click-based operator CLI (`lds-rag`) with ingest/index/search/answer/eval/stats commands. Unit tests cover the guardrail and tiering modules.",
  "tasks_to_mark_complete": [],
  "estimated_hours_remaining": 6
}