Support

Need help? We're here for you.

Contact Us

If you have questions, feedback, or need help with Prava Toolport, reach out:

We aim to respond within 48 hours.

Frequently Asked Questions

What exactly does Toolport do?

Toolport sits between your AI agent and all your MCP servers. Instead of every server loading its full tool list into the agent's context, the agent gets three small meta-tools — search_tools, get_tool_schema, and invoke_tool — and discovers the right tools on demand. Same power, a fraction of the context.

Which agents work with Toolport?

Any MCP-speaking client: Claude Code, Codex, Cursor, or your own agent. You register Toolport as a single MCP server in your agent's configuration, and it fronts every server you've added.

How do I add Toolport to Claude Code?

Add Toolport as an MCP server in your Claude Code configuration (one entry). Toolport announces the search → schema → invoke workflow to the agent automatically when it connects — no system-prompt changes needed.

Do my existing MCP servers need changes?

No. Toolport speaks standard MCP to your servers. Point it at the same server definitions you use today; it connects, asks each server for its tools, and indexes them. When a server's tools change, Toolport re-checks it automatically.

Does search work in languages other than English?

Yes. Toolport uses cross-lingual embeddings, so a query in one language finds tools described in another — for example, a Japanese query can find an English-described tool.

Does Toolport require an internet connection?

Semantic search and the index are fully local. You need network access only to reach remote MCP servers you've configured; local (stdio) servers and search itself work offline.

How does sign-in to hosted servers work?

For remote servers that require OAuth (for example, Microsoft or GitHub hosted services), Toolport opens the provider's sign-in in your browser and stores the tokens locally on your machine. Your credentials never pass through Prava Labs.

Which platforms and installs are there?

Same engine everywhere. On Mac (macOS 14+, Apple Silicon), the Mac App Store build sells the Solo tier billed by Apple — it is sandboxed, so it fronts remote (HTTP/OAuth) servers only; the direct download (notarized, unlocked with a license key) adds local (stdio) servers and carries Solo today, with Team and Enterprise coming soon. On Windows (10 & 11, native x64 & ARM64 — coming soon), the app is a free direct download that runs both local and remote servers at every tier, unlocked with the same license key. So local servers are available on every install except the sandboxed Mac App Store build.

I subscribed on the Mac App Store — can I switch to the direct download?

Yes, without paying twice:

  1. Cancel the subscription in System Settings → Apple Account → Subscriptions — Apple keeps it active until your renewal date, and we can't cancel it for you.
  2. Download the DMG edition whenever you like: it starts its own fresh 3-day trial, so you can set it up and confirm your local (stdio) servers work while your App Store subscription is still running.
  3. The two editions keep separate settings — you'll re-add your servers and OAuth sign-ins in the DMG edition.
  4. Email support with your Apple renewal date and we'll issue your Solo license key timed to start the day Apple's billing stops — no overlap, no gap in coverage.

What's the difference between the tiers?

Your tier unlocks components. Solo is the individual plan — semantic search, all platform modules, the first-party connectors, and your own MCP servers (local and remote, except on the sandboxed Mac App Store build). Team adds a shared team registry with governance and audit receipts, security pre-screening of tools, and token/cost metering. Enterprise adds air-gapped / on-prem deployment (offline license, no phone-home), SSO (OIDC), and audit export to your SIEM. See the plans overview.

How does per-seat licensing work?

Team and Enterprise are licensed per seat, where a seat is one user or agent — not a server and not a per-machine install. You run the registry on as many engine instances as you need for load or redundancy; servers are your infrastructure choice, not a licensing unit. For example, 20 users served by two load-balanced servers is 20 seats, not "two server licenses." Solo is a single seat for one person. (Prava Labs never sells per-server licenses — that would penalize running a redundant setup.)

The agent isn't finding my server's tools — what should I check?

The app is misbehaving — what should I do?