convez.ai
MCP ServerAutomaçãoAvançado
MrAk47Anand007

A360 MCP Server

MrAk47Anand007/A360-MCP

MCP para automação A360 com planejamento assistido e gerência de repositório

Ver no GitHub
InicialPluginScore · de 0 a 100

MCP server para Automation Anywhere A360 Control Room com suporte a operações de repositório, deployment e inteligência de pacotes. Integra APIs do Control Room com fluxos operacionais e auxilia na construção de bots assistida por IA, mantendo controle fino sobre validação e geração de JSON.

Instalação

$npx convez-ai MrAk47Anand007/A360-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
0
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Integrar Automation Anywhere A360 Control Room com agentes Claude
  • Automatizar criação, atualização e deployment de bots
  • Consultar pacotes disponíveis e resolver dependências de automações
  • Monitorar execução e atividades de bots em massa

Veredito

MCP robusto e bem arquitetado para A360, com CLI clara e suporte completo a auth/config. Recomendado para equipes que já utilizam Automation Anywhere e buscam integração profunda com Claude.

README

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

A360 MCP

a360-mcp is a package-first MCP server for Automation Anywhere A360 Control Room.

It keeps the clean npm/CLI shape from the nearby UiPath MCP project, but adapts it for A360 Control Room auth, repository operations, package intelligence, bulk workflows, and grounded AI-assisted bot building.

CLI

The package exposes the same primary CLI shape:

  • init
  • doctor
  • login
  • whoami
  • serve
  • logout

What This Repo Is For

This repo is designed to turn three things into one usable MCP:

  • live A360 Control Room APIs
  • your existing A360 operational workflows
  • natural-language driven bot planning and JSON generation

The important design choice is:

  • AI can help plan intent, choose packages, and sequence actions
  • final A360 JSON should be built by code, package metadata, templates, and validators
  • the MCP should expose both low-level APIs and high-value operational workflows

API Source Of Truth

This repo is grounded in exported Control Room swagger and live endpoint validation.

Primary exported specs:

  • swagger-export/swagger/api/v2/auth-api-supported.yaml
  • swagger-export/swagger/api/v2/repository-management-api.yaml
  • swagger-export/swagger/api/v3/deploy-api-supported.yaml
  • swagger-export/swagger/api/v3/bot-execution-orchestrator-api-supported.yaml
  • swagger-export/swagger/api/v2/packages-api-supported.yaml

Supporting generated inventory:

  • docs/a360-swagger-inventory.json
  • docs/control-room-editor-findings.md
  • docs/research/control-room-src/manifest.json

Additional implementation references were mined from:

  • ../a360_dataset
  • ../UiPath-to-A360-Migration

Quickstart

npm install
npm run extract:swagger
npm run extract:editor-sources
npm run login
npm run doctor
npm run serve

npm run login now supports an interactive path. If username, password, API key, or access token are missing from env/config, the CLI prompts for them and then persists the resulting access token in the package config store.

Configuration

Environment variables commonly used:

A360_BASE_URL=https://your-control-room-url
A360_AUTH_MODE=password
A360_USERNAME=your-user
A360_PASSWORD=your-password
A360_DEFAULT_FOLDER_ID=123456

Token mode is also supported:

A360_BASE_URL=https://your-control-room-url
A360_AUTH_MODE=token
A360_ACCESS_TOKEN=your-token
A360_DEFAULT_FOLDER_ID=123456

By default the package stores config under the current user's config directory. You can override the config path with A360_CONFIG_PATH.

Recommended auth pattern for MCP-first usage:

  1. npm run init
  2. npm run login
  3. npm run doctor
  4. npm run serve

This keeps Control Room auth inside the MCP package flow instead of relying on a separate browser session token.

Capability Areas

The MCP is organized into clear capability groups.

Repository

Low-level bot and repository operations:

  • a360_create_bot
  • a360_list_folder_items
  • a360_list_child_folders
  • a360_get_bot_content
  • a360_update_bot_content
  • a360_get_bot_dependencies
  • a360_update_bot_dependencies
  • a360_save_bot_bundle

Operations

Execution and deployment surfaces:

  • a360_deploy_automation
  • a360_list_activity
  • a360_list_recent_activity
  • a360_get_execution_details

Package Intelligence

Grounding layer for AI and package governance:

  • a360_list_available_packages
  • a360_get_package_versions
  • a360_get_package_command_schema
  • a360_resolve_package_metadata

Workflows

Bulk and folder-scoped operational tools:

  • a360_export_bots
  • a360_export_assets
  • a360_scan_package_usage
  • a360_plan_package_version_update
  • a360_apply_package_version_update
  • a360_scan_logtofile_issues
  • a360_apply_logtofile_fix
  • a360_apply_best_practice_scaffold
  • a360_silent_save_bot

AI Bot Generation

Prompt -> plan -> package metadata -> builder -> optional create:

  • a360_plan_bot_from_prompt
  • a360_build_bot_json_from_prompt
  • a360_create_bot_from_prompt

Validation

Preview, validate, and repair generated or edited bot JSON:

  • a360_validate_bot_json
  • a360_preview_bot_json
  • a360_normalize_bot_json
  • a360_fix_bot_json

Live UI capture tools

These tools build recorder steps from real Chrome state instead of guessed payloads. They connect to your Chrome over the DevTools Protocol. Start Chrome with:

chrome.exe --remote-debugging-port=9222 --user-data-dir=%TEMP%\a360-capture-profile

or set A360_MCP_CHROME_ENDPOINT to an existing DevTools URL.

ToolPurpose
a360_record_web_actionsExecute structured steps (navigate/click/type/select) live in Chrome; capture each target into canonical UIOBJECT/IMAGE payloads plus ready recorder nodes and suggested WINDOW variable scaffolding. Halts with ranked candidates on ambiguity.
a360_capture_ui_targetCapture one element (no action) into canonical payload pieces.
a360_validate_ui_target_bindingRe-capture a live target and compare it with a previously captured surrounding-context snapshot to estimate whether the binding is still reliable.
a360_repair_ui_target_bindingScan the live page for the best current match to a stored surrounding-context snapshot and return a patch-ready replacement payload plus ranked alternates.
a360_insert_recorder_stepInsert captured recorder node(s) and optional suggested variables into an existing bot and save via the normalized bundle flow.
a360_patch_step_targetReplace a node attribute's target payload with a captured one and save.
a360_repair_and_patch_ui_targetRepair a recorded target against the live page, patch the bot node, merge suggested variables, and save in one deterministic flow. If capturedContext is omitted, the tool first looks for persisted A360 MCP healing metadata already stored on the saved UIOBJECT.

The MCP client decomposes a natural-language prompt into structured steps; the server matches each target description deterministically against the page's elements — no AI guessing inside the server. Recorder command identity (packageName/commandName/attribute names) defaults to Recorder/Capture and is overridable per call; verify against a360_get_package_command_schema.

Recommended low-AI live capture pattern:

  1. Use AI only to reduce the prompt into a short structured action list.
  2. Use a360_record_web_actions or a360_capture_ui_target to capture real page targets.
  3. Use a360_get_package_command_schema to confirm exact Browser/Recorder attribute names.
  4. Use a360_insert_recorder_step or a360_save_bot_bundle to persist the bot.
  5. Validate with a360_preview_bot_json and a360_validate_bot_json before run/deploy.

The goal is to keep AI on intent extraction while MCP tools and package metadata own target capture, context validation, node shape, variable scaffolding, save, and validation.

Recorder targets captured by this MCP now persist a small a360Mcp metadata payload on the saved UIOBJECT typed value. That metadata stores the surrounding-context snapshot used for self-healing, so a future repair call can recover from live page drift without requiring the caller to resend the original captured context.

For post-build cleanup and standardization, use:

  • a360_apply_best_practice_scaffold to deterministically add Comment and LogToFile scaffolding plus required log path variables.

Current Tool Count

Current MCP tool count: 35

Build Flow For AI-Driven Bots

The intended layered flow is:

natural language prompt
-> intent extraction
-> package and command grounding
-> structured workflow plan
-> deterministic builder backend
-> validation / preview
-> optional create + save to Control Room

This avoids asking an LLM to invent raw A360 bot JSON directly.

The current planner now borrows the same major stages we already proved in the local UiPath-to-A360-Migration project:

  • package resolver style package ranking from live Control Room metadata
  • command/value context synthesis from normalized package commands and attributes
  • variable inference based on selected grounded commands
  • deterministic bot assembly through the TypeScript builder backend

Live-Tested Areas

The following surfaces have been validated live against a real Control Room tenant:

  • login and token persistence
  • folder listing
  • bot content reads
  • dependency reads
  • activity listing and execution detail lookup
  • create bot
  • update bot content
  • update bot dependencies
  • delete disposable bot
  • package list and package detail metadata lookup
  • prompt-driven dry-run build preview
  • bot JSON preview, validation, and structural fix flow

One known role-based limitation found during live testing:

  • package usage lookup can return 403 on /v2/packages/{name}/versions/usage depending on the tenant role

The package intelligence layer handles that gracefully and still resolves list/detail metadata.

Validation And Safety

The repo now supports preview-first behavior for higher-risk flows:

  • dryRun support for prompt-driven create flow
  • package-grounded JSON building
  • structural validation before save/create
  • automatic repair of small structural issues

Recommended pattern for AI-driven workflows:

  1. a360_plan_bot_from_prompt
  2. a360_build_bot_json_from_prompt
  3. a360_preview_bot_json
  4. a360_validate_bot_json
  5. a360_create_bot_from_prompt

Development

Useful commands:

npm test
npm run check
npm run build
npm run extract:swagger
npm run extract:editor-sources

Status

This repo is no longer just a scaffold. It now has:

  • working auth and repository primitives
  • working package intelligence
  • deterministic A360 JSON builder backend
  • prompt-driven planning/build/create flow
  • validation and repair tools

Next expected work is deeper planner coverage, richer command templates, and broader Control Room workflow support.

Ver o original no GitHub

Bom para

Relacionados