Field-by-field comparison
Compare the evidence before choosing an MCP.
Each signal stays separate, missing facts remain visible, and the column order follows your selection.
| Evidence field | mcp-agentWatchlist candidate | JADX-AI-MCP (Part of Zin MCP Suite)MCP100 Selected |
|---|---|---|
| Task fit | ||
| Best for |
|
|
| Not ideal for |
|
|
| Avoid when |
|
|
| Provenance | ||
| Provenance details | First-party MCP github:lastmile-ai · Publisher source ↗ | Community implementation github:zinja-coder · Publisher source ↗ |
| Maintenance | ||
| Maintenance details | Stale Repo: Not documented Package: Not documented | Repo: Aug 6, 2026 Package: Not documented |
| Popularity evidence | ||
| GitHub stars | Not documented | 2,664 GitHub stars · checked 2026-08-14T18:46:26.000Z |
| Client coverage | ||
| Client coverage details | Not documented | Claude Code✓ Local stdio Claude Desktop✓ Local stdio Codex✓ Local stdio VS Code✓ Local stdio Cursor✓ Local stdio OpenCode✓ Local stdio |
| Client coverage note | Local stdio runs on your computer. Streamable HTTP connects to a remotely hosted MCP server. | |
| Access and data | ||
| Authentication | Depends on configured MCP servers and LLM/cloud providers; the framework documents configuration/secrets and OAuth support. | The documented default is stdio with the HTTP endpoints on 127.0.0.1. The publisher explicitly warns that binding the MCP server to 0.0.0.0 or another non-localhost address exposes plain HTTP with no authentication or TLS, allowing anyone on that network to invoke all tools. |
| Cost | Open-source framework; model providers, Temporal/cloud deployment and connected services may be billable. | The publisher distributes the project under the Apache License 2.0 and lists no MCP server usage fee. JADX, local compute, the selected MCP client, and model-provider costs can still apply. |
| Permissions | Inherits the tool/resource authority of every connected MCP server plus any workflow/cloud runtime permissions. | The published tools can read decompiled code for classes and methods, smali, AndroidManifest.xml, string and resource files, cross-references, and debugger frames, threads, and variables from the project open in JADX. They can also rename classes, methods, fields, packages, and variables within the JADX analysis workspace. |
| Data handling | Workflow data can pass through connected MCP servers, model providers and optional cloud/Temporal services; deployment-specific policies control retention. | The publisher's architecture sends tool requests from the LLM client through the Python MCP server over HTTP to the JADX-GUI plugin, then returns source, resource, manifest, and debugger data to the connected client. The reviewed README does not document telemetry, retention, or model-training use, so those behaviors remain unknown. |
| Limitations | ||
| Tradeoffs |
|
|
| Risk context | Broad authority across multiple MCP servers/providers makes permission and production-governance evidence important before Selected promotion. | This server exposes sensitive reverse-engineering material and mutation tools for the APK project open in JADX. Keep both connections on 127.0.0.1 and prefer stdio; if remote access is unavoidable, use a trusted isolated network plus firewall or SSH tunnel. Treat decompiled source, resources, manifests, and debugger values as potentially confidential, and review rename operations before saving or exporting analysis artifacts. |
| Evidence date | ||
| Editorial review | 2026-09-13T00:00:00Z | 2026-08-11 |
| Candidate evidence | 2026-09-13T00:00:00Z | 2026-08-14T18:46:26.000Z |
Popularity, maintenance, fit, permissions, and client support are independent evidence fields.
Read the evidence method →