> ## Documentation Index
> Fetch the complete documentation index at: https://docs.aisim.sandboxaq.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Platform Compatibility
> Which MCP clients SandboxAQ has validated against the LQM tools, what works on each, and the known limitations.
## What "supported" means here
Support is defined **per platform, per tool, and per feature**. This page lists only the platforms on which SandboxAQ has validated tool calling and documented the known limitations.
All LQM tools on the [SandboxAQ MCP Server](/mcp-server/mcp-server) are generally compatible with any client that supports MCP tool calling and — where the tool requires it — file uploads. **Platforms not listed here should not be assumed supported.** They may work; SandboxAQ has not tested them.
Looking for setup steps rather than a support matrix? See the [Quickstart](/getting-started/quickstart) for the condensed version, or [Connect your MCP client](/getting-started/connect-client) for the full walkthrough.
## Tool coverage
Which of the [five SandboxAQ tools](/mcp-server/mcp-server#available-tools) have been validated on each platform.
| Tool | Claude Desktop | claude.ai | Claude Code | Claude Science | Antigravity |
| ------------------------------------------------- | -------------- | --------- | ----------- | -------------- | ----------- |
| [`aqcat`](/models/aqcat) | ✅ | ✅ | ✅ | ✅ | ✅ |
| [`AqpotencyPotency`](/models/aqpotency#tools) | ✅ | ✅ | ✅ | ✅ | ✅ |
| [`AqpotencyScreen`](/models/aqpotency#tools) | ✅ | ✅ | ✅ | ❌ | ✅ |
| [`AqpotencyScan`](/models/aqpotency#tools) | ✅ | ✅ | ✅ | ⚠️ | ✅ |
| [`AqpotencySelectivity`](/models/aqpotency#tools) | ✅ | ✅ | ✅ | ✅ | ✅ |
**Legend:** ✅ validated · ⚠️ invokes, but not reliably · ❌ confirmed not supported
**On Claude Science tool invocation is inconsistent.** A call that fires correctly once may not the next time. Name the tool explicitly in your prompt and supply every required parameter up front — see [Improving tool-invocation reliability](/mcp-server/mcp-server#improving-tool-invocation-reliability).
## Platform capabilities
Capabilities that apply to a whole platform rather than to individual tools. Which tools work where is in [Tool coverage](#tool-coverage) above.
| Platform | Tested version | File upload | Result card |
| :---------------------- | :------------------- | :---------- | :---------- |
| **Claude Desktop app** | `1.24012.9 (03c61d)` | ✅ | ✅ |
| **claude.ai (web)** | — | ✅ | ✅ |
| **Claude Code (CLI)** | — | ✅ | N/A |
| **Claude Science** | `0.1.27` | ❌ | ❌ |
| **Antigravity Desktop** | `2.8.0` | ✅ | ❌ |
The two columns mean:
* **File upload** — the platform can push a compound library through the upload-session flow (`upload_files` / `upload_local_files`) that `AqpotencyScreen` depends on. Without it, `AqpotencyScreen` cannot run.
* **Result card** — the platform renders the tool's structured output as a formatted result card rather than plain text. Marked **N/A** for Claude Code, whose terminal interface is text-only by design.
**`AqpotencyScreen` is unavailable wherever file upload is unsupported.** The tool takes an `upload_session_id`, not an inline compound list, so a platform that cannot upload a `.smi` or `.csv` library cannot call it at all. On Claude Science, use `AqpotencyScan` (one ligand against a curated panel) or run the screen from Claude Desktop, claude.ai, or Claude Code instead.
## Platform notes
The reference platforms. All five tools call reliably, library uploads work, and structured results render as cards.
Connectors are stored on your Claude account, so configuring one configures both. Setup: [Connect your MCP client → Claude Desktop and claude.ai](/getting-started/connect-client).
All five tools call reliably and file uploads work. There is no result card because the CLI renders text, not UI components — you get the same structured JSON, formatted by Claude in the transcript.
Setup: [Connect your MCP client → Claude Code](/getting-started/connect-client).
Tool invocation is **inconsistent**: `aqcat`, `AqpotencyPotency`, `AqpotencyScan`, and `AqpotencySelectivity` succeed some of the time but cannot be relied on. File upload is not supported, so `AqpotencyScreen` is unavailable, and results are returned as text rather than result cards.
Claude Science inherits its connectors from your Claude account, so the SandboxAQ connection carries over automatically once Claude Desktop or claude.ai is set up — and so does the **Lazy load tools** setting, which must be changed on your Claude account rather than inside Claude Science. Setup: [Connect your MCP client → Claude Science](/getting-started/connect-client).
If a call doesn't fire, name the tool explicitly in your prompt and supply every required parameter up front. See [Improving tool-invocation reliability](/mcp-server/mcp-server#improving-tool-invocation-reliability).
All five tools call reliably and file uploads work. There is no result card because they are currently unsuported.
Antigravity uses `serverUrl` for HTTP MCP servers — **not** `url`, which is what the Claude clients use. Config details: [SandboxAQ MCP Server → Antigravity](/mcp-server/mcp-server#antigravity).
## Platforms not listed
Any client that speaks the Model Context Protocol over Streamable HTTP can connect to `https://mcp.platform.aisim.sandboxaq.com` using the same OAuth flow. Those clients are **untested**, not unsupported: SandboxAQ has not validated tool calling or documented their limitations, so treat any behaviour there as unverified.
If you need a platform validated, tell us which one and how you use it — [support.aisim@sandboxaq.com](mailto:support.aisim@sandboxaq.com), or use the [contact form](/support-legal/contact). Enterprise customers with a platform requirement: [mcp-prod@sandboxaq.com](mailto:mcp-prod@sandboxaq.com).
## Reporting an issue
Found a platform-specific bug, or a tool that won't fire on a client listed above? Email [support.aisim@sandboxaq.com](mailto:support.aisim@sandboxaq.com) or use the [contact form](/support-legal/contact). Include your client and version, the tool you called, and the prompt you used — the version column above is the build each platform was validated against, so a mismatch is worth mentioning.
This is the standard support channel for the **Individual tier**, and it's best effort. Enterprise customers have separate terms and a named contact — see [Enterprise Tier Support](/support-legal/enterprise-tier-support).
## Next steps
Full setup walkthroughs, client-specific caveats, and troubleshooting.
How the server works, the full tool list, limits and quotas, and the security model.
Client settings and prompt phrasing that make tool calls fire consistently.
Short answers on supported clients, account setup, accuracy, privacy, and billing.