Pod

Yes, this is on purpose. Pod is built for agents, so the default page is plain HTML.
Human? View the normal website.
Agent? You probably prefer /mcp/devplan-mcp-server/issues.md or /mcp/devplan-mcp-server/issues.json, or Pod over MCP.

Reported issues for DevPlan MCP Server

Pod holds 17 of 47 problems reported by people outside the maintainer team. Issues filed by the project's own owners, members and collaborators are excluded entirely — a maintainer's release checklist is not a warning to a prospective user.

Back to DevPlan MCP Server.

Most discussed

bug: Feature phases treat individual API endpoints as separate phases (13 phases for 7 features)

Description

The generated plan creates 13 separate feature phases (Phase 2 through Phase 13) by treating each bullet from the MVP features list as its own phase. This includes individual API endpoints as standalone phases:

Read the thread · 2026-02-14 · closed · 3 comments

bug: Plan scaffold generates nonsensical file paths (.txt extension, Python naming for Java)

Description

The generated plan scaffold creates file paths that don't match the detected language conventions.

Reproduction

Generate a plan for a Java 21 + Spring Boot 3 project.

Expected

Java-style paths:

Actual

Read the thread · 2026-02-14 · closed · 3 comments

bug: CLAUDE.md and agent generators ignore detected language, emit Python/TypeScript defaults

Description

The `devplan_generate_claude_md`, `devplan_generate_executor`, and `devplan_generate_verifier` tools all ignore the detected language from the brief and emit hardcoded Python/TypeScript/Node.js defaults instead.

Reproduction

  1. Create a PROJECT_BRIEF.md for Java 21 + Spring Boot 3 + Gradle + JUnit 5
  2. Call `devplan_generate_claude_md` with `language: "java"`
  3. Call `devplan_generate_executor` with `language: "java"`
  4. Call `devplan_generate_verifier` w

Read the thread · 2026-02-14 · closed · 3 comments

bug: Feature grouping over-corrects — 6 of 7 features dropped, only Leaderboard remains

Description

The fix for #143 (endpoints listed as separate phases) over-corrected. Instead of grouping 13 items into ~6 logical feature phases, it collapsed everything into 1 phase and lost 6 of 7 features entirely.

Reproduction

Generate a plan for corporate-buzzword-bingo with these 7 features + 5 endpoints in the brief:

  1. Buzzword database with categories and difficulty ratings
  2. Bingo card generator (unique 5x5 cards, FREE space center)
  3. Meeting session management (create

Read the thread · 2026-02-14 · closed · 2 comments

🟡 P1: v2/nice-to-have features get weird phase numbers (14.5, 15.5, etc.)

Problem

Nice-to-have features from the brief become phases with decimal numbers that look broken:

## Phase 14.5 (v2): Prometheus metrics
## Phase 15.5 (v2): Hybrid search
## Phase 16.5 (v2): Embedding caching
## Phase 17.5 (v2): Quantized INT8 model

Expected Behavior

Nice-to-have features should be grouped cleanly:

Option A - Single deferred section:

## Future Enhancements (v2)

These features are deferred until after MVP:

### Prometheus metrics
- Requirem

[Read the thread](https://github.com/mmorris35/devplan-mcp-server/issues/128) · 2026-02-06 · closed · 2 comments

### bug: Duplicate subtask ID 1.1.2 in generated plan

## Description

The generated plan scaffold contains two different subtasks both numbered 1.1.2:

1. **Subtask 1.1.2: API Endpoints Implementation** — implements the 5 API endpoints from domain specs
2. **Subtask 1.1.2: First Core Feature** — implements the first core feature

## Expected

Each subtask should have a unique ID (e.g., 1.1.2 and 1.1.3).

## Root Cause

The domain spec endpoint extraction (from #126 fix) inserts a subtask for API endpoints, but the existing scaffold also generates a

[Read the thread](https://github.com/mmorris35/devplan-mcp-server/issues/141) · 2026-02-14 · closed · 1 comment

### bug: Tech Stack shows PostgreSQL instead of H2 from brief constraints

## Description

When the PROJECT_BRIEF.md explicitly specifies `H2 in-memory database` in the Must Use section, the generated plan's Technology Stack section shows `Database: PostgreSQL` instead.

## Reproduction

1. Create a PROJECT_BRIEF.md with `Must Use: H2 in-memory database`
2. Run `devplan_generate_plan`
3. Check the Technology Stack section

## Expected


## Actual


## Root Cause

The database field in the Technology Stack ap

[Read the thread](https://github.com/mmorris35/devplan-mcp-server/issues/138) · 2026-02-14 · closed · 1 comment

### bug: "unknown" language placeholder leaks into plan scaffold text

## Description

When generating a plan for a Java/Spring Boot project, the language is correctly detected in the Technology Stack section (`Language: Java, Framework: Spring Boot`), but the detected language does not propagate into the scaffold body text.

## Reproduction

1. Create a PROJECT_BRIEF.md specifying Java 21, Spring Boot 3, Gradle, JUnit 5
2. Run `devplan_generate_plan` with the brief
3. Observe the scaffold output

## Expected

Phase 0 text should reference Java conventions, Gradle 

[Read the thread](https://github.com/mmorris35/devplan-mcp-server/issues/137) · 2026-02-14 · closed · 1 comment

## Most recent

### [Revised] Feature: Skip feature phase scaffold - let Opus handle it

## Problem: Feature phase scaffold is too generic, Opus throws it out and starts from scratch

**Severity:** High

**User Report:**
> Currently it generates scaffolds that are nowhere close to the end result. Opus has to throw out the scaffold and start from scratch every time. Hell, just dropping the scaffold step would be an improvement at this point.

### Root Cause Analysis

In `src/generators.ts`, the function `generateFeaturePhases()` produces feature phases that:

1. **Claim to be complet

[Read the thread](https://github.com/mmorris35/devplan-mcp-server/issues/159) · 2026-03-21 · closed · 0 comments

### Feature: Add option to skip feature phase scaffold generation

## Problem: Feature phase scaffold is too generic, Opus throws it out and starts from scratch

**Severity:** High

**User Report:**
> Currently it generates scaffolds that are nowhere close to the end result. Opus has to throw out the scaffold and start from scratch every time. Hell, just dropping the scaffold step would be an improvement at this point.

### Root Cause Analysis

In `src/generators.ts`, the function `generateFeaturePhases()` produces feature phases that:

1. **Claim to be complet

[Read the thread](https://github.com/mmorris35/devplan-mcp-server/issues/158) · 2026-03-21 · closed · 0 comments

### Add execution suggestion after validation phase

After DevPlan completes the validation phase, the output should explicitly suggest using Haiku as the executor.

**Suggested output:**
> **Plan ready for execution.** Exit this session and start a fresh one with Haiku loaded, then ask Claude to implement the phases.

**Why:**
- Formalizes the optimal workflow discovered through trial-and-error
- Makes the Haiku executor pattern discoverable
- Users shouldn't have to learn the hard way that Opus overthinkes simple tasks

**Context:**
- Using Haik

[Read the thread](https://github.com/mmorris35/devplan-mcp-server/issues/157) · 2026-03-11 · open · 0 comments

### emit scripts with the plan

What's happening now: Each subtask in a DevPlan (like "1.2.3: Implement auth middleware") gets sent to Haiku as a natural language prompt. Haiku reads the full context (CLAUDE.md + DEVELOPMENT_PLAN.md), interprets what to do, and generates/executes code. That's a lot of tokens burned on tasks where the what to do is deterministic even if the how to implement requires LLM reasoning.
Where shell scripts would be a clear win — the "mechanical" subtasks:
Many subtasks in a typical DevPlan are essent

[Read the thread](https://github.com/mmorris35/devplan-mcp-server/issues/156) · 2026-03-04 · open · 0 comments

### research: Ideas from Anthropic's official Skills guide

## Source

Anthropic's official "Complete Guide to Building Skills for Claude" (30 pages)
PDF: https://resources.anthropic.com/hubfs/The-Complete-Guide-to-Building-Skill-for-Claude.pdf

## Relevant Patterns

### Progressive Disclosure (3 levels)
1. **YAML frontmatter** — Always loaded, minimal, helps Claude decide relevance
2. **SKILL.md body** — Loaded when triggered
3. **Linked files** — Loaded only as needed

**DevPlan application:** Could executor/verifier agents use this pattern? Frontmatte

[Read the thread](https://github.com/mmorris35/devplan-mcp-server/issues/152) · 2026-02-16 · open · 0 comments

### feat: Add --checkpoint-between-waves flag to devplan_generate_executor

## Summary

Add an optional flag to `devplan_generate_executor` that injects `/clear` + re-read instructions between waves in the generated executor agent.

## Motivation

Based on Ashley Ha's "subtractive context management" methodology (UC Berkeley ML, prev. Rivian):
- Never exceed 60% of context window
- `/clear` between phases prevents instruction drift
- Fresh context = Claude actually reads instructions with full attention

Thread with 5.2K views validating this approach: https://www.threa

[Read the thread](https://github.com/mmorris35/devplan-mcp-server/issues/151) · 2026-02-16 · open · 0 comments

### Subtasks need inline completion instructions for Haiku agent

## Problem
The completion marking instructions (how to check off subtasks when done) are currently only in CLAUDE.md. Since the Haiku agent can only concentrate on what's immediately in front of it, it doesn't follow these instructions.

## Current Behavior
- Completion instructions live in CLAUDE.md
- Haiku agent executes subtasks from DEVELOPMENT_PLAN.md
- Agent doesn't read linked documents or remember external instructions
- Subtasks don't get marked complete

## Expected Behavior
Each subta

[Read the thread](https://github.com/mmorris35/devplan-mcp-server/issues/148) · 2026-02-14 · open · 0 comments

### feat: Local LLM support via Ollama for plan execution

## Summary

Add support for using local LLMs (via Ollama) as an alternative to Claude Haiku for executing development plans. This enables:
- **Zero API costs** after hardware investment
- **Privacy** — code never leaves the machine
- **Offline execution** — no internet required

## Motivation

Models like **Qwen3-Coder-Next-80B** now rival Claude on coding benchmarks and can run locally on Apple Silicon Macs with 64GB+ unified memory. For teams with suitable hardware, this eliminates per-token c

[Read the thread](https://github.com/mmorris35/devplan-mcp-server/issues/145) · 2026-02-14 · open · 0 comments

### Add CHANGELOG.md as a generated output artifact

## Feature Request

Add a **CHANGELOG.md** (or CHANGES.md) as one of the standard output documents generated by DevPlan MCP during project startup.

## Rationale

When working on a project over time, changes accumulate but often aren't documented until release. Having a change management document from day one:

1. **Tracks what was built and why** - Beyond just git commits, captures the intent and context
2. **Documents post-MVP changes** - Like the "Post-MVP Implementation Record" section we ma

[Read the thread](https://github.com/mmorris35/devplan-mcp-server/issues/135) · 2026-02-10 · open · 0 comments

The remaining reports are on [the project's issue tracker](https://github.com/mmorris35/devplan-mcp-server/issues).