Skip to main content
Glama

@mudravaorg/mcp-server

MCP (Model Context Protocol) server for PixelFixer — a visual bug tracking and kanban board tool. Connects AI agents (Claude, Cursor, VS Code Copilot, Windsurf) to your PixelFixer projects.

What it does

This MCP server gives AI agents full access to your PixelFixer workspace:

Category

Tools

Session

init_session, set_context

Teams & Projects

list_teams, list_projects, get_project, list_team_members

Tasks

list_tasks, get_task, create_task, update_task, move_task, search_tasks

Comments

add_comment, list_comments

Kanban

list_columns

GitHub

get_github_context, get_repo_tree, get_file_content, create_pull_request, commit_files

AI Pipeline

start_task, complete_ai_task, list_ai_queue

v1.0 highlights

  • Session context — call init_session once; all tools auto-use your team/project IDs

  • Task numbers — get_task, start_task, etc. accept taskNumber: 42 instead of raw IDs

  • Compact responses — list views return summaries (~90% fewer tokens than v0.2)

  • Retry & timeout — automatic retry with backoff on 429/5xx, 30s request timeout

  • Better errors — human-readable messages that help AI self-correct

Related MCP server: kanban-mcp

Quick Start

1. Get an API Token

Go to PixelFixer → Team Settings → API Tokens and create a token with read + write scopes.

2. Configure your IDE

VS Code (.vscode/mcp.json):

{
  "servers": {
    "pixelfixer": {
      "command": "npx",
      "args": ["-y", "@mudravaorg/mcp-server"],
      "env": {
        "PIXELFIXER_API_TOKEN": "pf_your_token_here",
        "PIXELFIXER_API_URL": "https://pixelfixer.mudrava.com"
      }
    }
  }
}

Cursor (.cursor/mcp.json) / Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "pixelfixer": {
      "command": "npx",
      "args": ["-y", "@mudravaorg/mcp-server"],
      "env": {
        "PIXELFIXER_API_TOKEN": "pf_your_token_here",
        "PIXELFIXER_API_URL": "https://pixelfixer.mudrava.com"
      }
    }
  }
}

Note: VS Code uses "servers" as the root key, while Cursor and Claude Desktop use "mcpServers".

3. Start using it

Ask your AI agent:

  • "Check my PixelFixer tasks and fix what's in the AI queue"

  • "Start task #42 and fix it"

  • "Search for high-priority open bugs"

  • "Create a PR that fixes the button color issue from task #15"

Environment Variables

Variable

Required

Default

Description

PIXELFIXER_API_TOKEN

Yes

—

Personal API token (starts with pf_)

PIXELFIXER_API_URL

No

http://localhost:3000

PixelFixer instance URL

Local Development

If you're running PixelFixer from source, you can use the local build instead of the npm package:

cd packages/mcp-server
pnpm install
pnpm build

Then point your IDE to the local file:

{
  "servers": {
    "pixelfixer": {
      "command": "node",
      "args": ["${workspaceFolder}/packages/mcp-server/dist/index.js"],
      "env": {
        "PIXELFIXER_API_TOKEN": "pf_your_token_here",
        "PIXELFIXER_API_URL": "http://localhost:3000"
      }
    }
  }
}

Tool Reference

init_session

The recommended first call in every session. Auto-discovers your team and project (if you have exactly one of each), sets the session context, and returns the AI task queue as compact summaries. After this, all tools auto-fill teamId/projectId.

start_task

Start working on an AI task. Moves the task to In Progress, sets AI status to PROCESSING, and returns full context: task details, comments, GitHub info, columns, and a workflow guide. Accepts taskId or taskNumber.

search_tasks

Search with multiple filters:

  • q — text search (title, description, task number)

  • status — OPEN, IN_PROGRESS, RESOLVED, CLOSED

  • priority — LOW, MEDIUM, HIGH, CRITICAL

  • aiStatus — NONE, QUEUED, PROCESSING, COMPLETED, FAILED

  • assigneeId, columnId, tag

create_pull_request

Creates a branch and PR in the connected GitHub repo. Example:

branchName: "fix/PF-42-button-color"
title: "Fix button color on dashboard"
body: "Resolves PF-42. Changed primary button color..."

Security

  • API tokens are hashed (SHA-256) in the database — even a DB leak won't expose tokens

  • Tokens have scoped permissions (read / write / admin)

  • The MCP server runs locally on your machine — data goes directly to your PixelFixer instance over HTTPS

  • No data is sent to third parties

License

MIT

Available Tools

26 tools
add_commentB

Add a comment to a task. Call AFTER making changes and BEFORE complete_ai_task. IMPORTANT: content in English.

Inputs: content (required), taskId, teamId, projectId, +1 more.

batch_complete_tasksA

Complete multiple AI tasks in a single call. Each entry can have its own message, prUrl, and commitHash. Status defaults to COMPLETED.

Inputs: tasks (required), teamId, projectId.

batch_get_tasksA

Get full details of multiple tasks in a single call. Accepts an array of task numbers. More efficient than calling get_task multiple times.

Inputs: taskNumbers (required), teamId, projectId.

batch_start_tasksA

Start multiple AI tasks at once. Sets each to IN_PROGRESS / PROCESSING. Returns compact results. More efficient than calling start_task multiple times.

Inputs: taskNumbers (required), teamId, projectId.

commit_filesA

Commit files to a branch. Call AFTER create_pull_request. Max 50 files. IMPORTANT: commit message in English.

Inputs: files (required), branch (required), message (required), teamId, +1 more.

complete_ai_taskA

Report the result of AI work. Auto-moves task to Review column. Call AFTER add_comment. IMPORTANT: message in English. Status defaults to COMPLETED if omitted.

Inputs: prUrl, status, taskId, teamId, +4 more.

create_pull_requestA

Create a branch and pull request. Call BEFORE commit_files. IMPORTANT: all text in English.

Inputs: title (required), branchName (required), body, teamId, +2 more.

create_taskA

Create a new task. Requires title and columnId. IMPORTANT: title and description MUST be in English.

Inputs: title (required), columnId (required), tags, teamId, +5 more.

get_file_contentA

Read a file from the connected GitHub repository.

Inputs: path (required), ref, teamId, projectId.

get_github_contextB

Get the GitHub repository connection for a project. Returns null if no repo is connected.

Inputs: teamId, projectId.

get_projectB

Get details of a specific project including its GitHub connection.

Inputs: teamId, projectId.

get_repo_treeC

Browse the file tree of the connected GitHub repository.

Inputs: ref, path, teamId, projectId.

get_taskA

Get full details of a task: description, metadata, screenshot, page URL, CSS selector, browser info, console/network errors, comments, AI status. Accepts taskId OR taskNumber.

Inputs: taskId, teamId, projectId, taskNumber.

init_sessionA

Initialize the MCP session. Call this FIRST before any other tool. Auto-discovers your team/project from the API token and returns the AI task queue. If you have exactly one team and one project, they are auto-selected for all subsequent calls (no need to pass teamId/projectId). If you have multiple teams or projects, call set_context afterward to select which one to work with.

Inputs: no named parameters.

list_ai_queueA

List tasks queued for AI processing (compact summaries). Supports optional priority and tag filters. Use start_task on each to begin work.

Inputs: tag, teamId, priority, projectId.

list_columnsB

List all kanban columns in a project.

Inputs: teamId, projectId.

list_commentsA

List all comments on a task.

Inputs: taskId, teamId, projectId, taskNumber.

list_projectsB

List all projects in a team.

Inputs: teamId.

list_tasksA

List all tasks in a project (compact summaries). Use get_task for full details of a specific task.

Inputs: teamId, projectId.

list_team_membersB

List all members of a team.

Inputs: teamId.

list_teamsA

List all teams the authenticated user belongs to.

Inputs: no named parameters.

move_taskA

Move a task to a different kanban column. Only use when explicitly asked — start_task and complete_ai_task handle column moves automatically.

Inputs: columnId (required), taskId, teamId, position, +2 more.

search_tasksA

Search tasks with filters. Returns compact summaries. Use get_task for full details.

Inputs: q, tag, limit, status, +6 more.

set_contextA

Set or change the active team and/or project for this session. Use this after init_session when you have multiple teams/projects. Once set, all subsequent tool calls will use these IDs automatically — no need to pass teamId/projectId every time.

Inputs: teamId, projectId.

start_taskA

Start working on an AI task. Moves task to In Progress, sets AI status to PROCESSING, returns full context (task, comments, GitHub info, columns). Accepts taskId OR taskNumber. Call this FIRST for every AI task.

Inputs: taskId, teamId, projectId, taskNumber.

update_taskA

Update task properties. Do NOT use this to move columns — use start_task/complete_ai_task instead. Accepts taskId OR taskNumber. IMPORTANT: title/description in English.

Inputs: title, status, taskId, teamId, +6 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.5/5.0

    Scored across 26 tools

    Disambiguation4/5

    Most tools map to distinct actions/resources, but get_project's description already includes GitHub connection info, overlapping with get_github_context; list_tasks and list_ai_queue also require careful reading. The batch variants are clearly labeled as bulk versions, reducing confusion.

    Naming Consistency5/5

    All tools use a consistent lowercase snake_case verb_noun pattern such as init_session, list_tasks, complete_ai_task, and batch_get_tasks. The pattern is predictable across session, task, comment, and GitHub operations.

    Tool Count2/5

    At 26 tools, the server exceeds the 25-tool threshold and feels heavy. Several batch_* variants add redundant surface area that could be handled with optional parameters on the singular tools.

    Completeness4/5

    The core lifecycle is well covered: session/context setup, task CRUD and search, comments, kanban states, AI queue handling, GitHub file browsing, and PR creation/commit. Minor gaps such as deleting tasks, updating comments, or checking PR status are workarounds rather than dead ends.

    Maintenance

    ActivityStale
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      B
      quality
      D
      maintenance
      An MCP server that provides a database-backed kanban board with 40+ tools for AI agents to track issues, features, todos, epics, and diary entries across projects, including status workflows, relationships, and semantic search.
      45
      86
      MIT