Arm MCP Server
arm/mcp
Ferramentas Arm para migração x86→Arm e otimização de performance
Ver no GitHubServidor MCP especializado para desenvolvimento, migração e otimização em arquitetura Arm. Fornece ferramentas para análise de compatibilidade de código, busca em base de conhecimento Arm, inspeção de arquiteturas de containers e análise de performance de assembly.
Instalação
npx convez-ai arm/mcp1 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
Casos de uso
- Migração de aplicações x86 para Arm
- Análise de compatibilidade de código com arquitetura Arm
- Busca semântica em documentação e recursos Arm
- Inspeção de suporte de arquitetura em imagens Docker
- Análise de performance de código assembly com LLVM-MCA
Veredito

Ferramenta robusta e bem mantida (certificação OpenSSF e SLSA Level 3) para equipes que precisam migrar ou otimizar workloads em arquitetura Arm. Integração direta com Docker e suporte a múltiplos assistentes de IA tornam-na prática para pipelines de CI/CD e desenvolvimento local.
README
A tradução para o português deste README ainda está na fila.
Arm MCP Server
An MCP server providing AI assistants with tools and knowledge for Arm architecture development, migration, and optimization.
Using the Arm MCP Server
If your goal is to migrate an application from x86 to Arm as quickly as possible, start here:
Automate x86-to-Arm application migration using Arm MCP Server
Features
This MCP server equips AI assistants with specialized tools for Arm development:
- Knowledge Base Search: Keyword and semantic search across Arm documentation, learning resources, intrinsics, and software compatibility information, with technical-name aliases
- Code Migration Analysis: Scan codebases for Arm compatibility using migrate-ease (supports C++, Python, Go, JavaScript, Java)
- Container Architecture Inspection: Check Docker image architecture support using integrated Skopeo and check-image tools.
- Assembly Performance Analysis: Analyze assembly code performance using LLVM-MCA
Pre-Built Image
If you would prefer to use a pre-built, multi-arch image, the official image can be found in Docker Hub here: armlimited/arm-mcp:latest
Prerequisites
- Docker with Buildx support
- An MCP-compatible AI assistant client (e.g. GitHub Copilot, Kiro CLI, Codex CLI, Claude Code, etc)
Quick Start
1. Build the Docker Image
From the root of this repository:
docker buildx build -f mcp-local/Dockerfile -t armlimited/arm-mcp . --load
This builds for the Docker host's native architecture. The release workflow is responsible for explicit multi-architecture builds.
For local development builds, see the Quick Local Development Builds section in CONTRIBUTING.md.
2. Configure Your MCP Client
Choose the configuration that matches your MCP client:
Claude Code
Add to .mcp.json in your project:
{
"mcpServers": {
"arm-mcp": {
"command": "docker",
"args": [
"run",
"--rm",
"-i",
"--pull=always",
"-v", "/path/to/your/workspace:/workspace",
"armlimited/arm-mcp"
]
}
}
}
GitHub Copilot (VS Code)
Add to .vscode/mcp.json in your project, or globally at ~/Library/Application Support/Code/User/mcp.json (macOS):
{
"servers": {
"arm-mcp": {
"type": "stdio",
"command": "docker",
"args": [
"run",
"--rm",
"-i",
"--pull=always",
"-v", "/path/to/your/workspace:/workspace",
"armlimited/arm-mcp"
]
}
}
}
The easiest way to open this file in VS Code for editing is command+shift+p and search for
MCP: Open User Configuration
AWS Kiro CLI
Add to ~/.kiro/settings/mcp.json:
{
"mcpServers": {
"arm-mcp": {
"command": "docker",
"args": [
"run",
"--rm",
"-i",
"--pull=always",
"-v", "/path/to/your/workspace:/workspace",
"armlimited/arm-mcp"
],
"timeout": 60000
}
}
}
Gemini CLI
It is recommended to use a project-local configuration file to ensure the relevant workspace is mounted.
Add to .gemini/settings.json in your project root:
{
"mcpServers": {
"arm-mcp": {
"command": "docker",
"args": [
"run",
"--rm",
"-i",
"--pull=always",
"-v", "/path/to/your/workspace:/workspace",
"armlimited/arm-mcp"
]
}
}
}
MCP Clients using TOML format (e.g. Codex CLI)
[mcp_servers.arm-mcp]
command = "docker"
args = [
"run",
"--rm",
"-i",
"--pull=always",
"-v", "/path/to/your/workspace:/workspace",
"armlimited/arm-mcp"
]
Note: Replace /path/to/your/workspace with the actual path to your project directory that you want the MCP server to access.
3. Restart Your MCP Client
After updating the configuration, restart your MCP client to load the Arm MCP server.
Knowledge-base search
knowledge_base_search returns up to five ranked chunks by default. Optional settings:
k: an integer from 1 to 20 limiting initial matches, defaulting to 5 when omitted. Explicitnulland values outside this range are rejected.filters:doc_type,product,platform, oredition, each accepting a label or non-empty list. Values within a field use OR; fields use AND. Matching ignores case and surrounding whitespace. Use known metadata labels; records with missing tags cannot match a filter.
{"query": "How do SVE predicated loads work?", "filters": {"doc_type": "Intrinsic", "product": "SVE"}, "k": 3}
Choose scope for the amount of context:
| Scope | Content returned |
|---|---|
chunks (default) | Ranked matching chunks. |
page | All stored chunks from each matched page, in order. |
Page expansion returns shared pages once. Filters and k select initial matches;
they do not limit expanded content. Use expansion only when the user requests
wider context, since responses can be large. No live pages are fetched.
Expansion requires regenerated KB artifacts with complete ordering metadata;
missing or conflicting context returns an error. The shared search library retains
parent_resource; it is temporarily disabled in MCP pending further evaluation.
See the shared search documentation for API and metadata details.
Logging
Depending on usage, the server may write two log files under /workspace. With the
configuration examples above, these files appear in the project directory on
your computer:
mcp-traffic.jsonlrecords when tools are used, the inputs provided, and the reason for each tool call. It also records results from knowledge base searches.error_logging.yamlrecords details about errors encountered by the server. This information can help with troubleshooting.
These logs may contain information from your project and tool requests. Review their contents before sharing them.
Development and contributing
See CONTRIBUTING.md for repository structure, integration testing, reproducible build inputs, runtime-egress validation, dependency updates, and release workflows.
Troubleshooting
Accessing the Container Shell
To debug or explore the container environment:
docker run --rm -it --entrypoint /bin/bash armlimited/arm-mcp
Common Issues
- Timeout errors during migration scans: Increase the
timeoutvalue in your MCP client configuration (e.g.,"timeout": 120000for 2 minutes) - Empty workspace: Ensure your volume mount path is correct and the directory exists
- Architecture mismatches: Confirm that the local image matches the Docker host's native architecture; use the release workflow for explicit cross-platform builds.
Suspected vulnerabilities must not be reported in public issues. See the Arm MCP security policy and report them via Arm PSIRT.
License
Copyright © 2026, Arm Limited and Contributors. All rights reserved.
Licensed under the Apache License, Version 2.0. See LICENSE for details.