Q & A

Is MCPBus open source?

The core MCPBus binary is proprietary, but it stands on the shoulders of giants. We proudly use open-source packages. You can view the full list of third-party licenses directly in your terminal at any time by running:

mcpbus --licenses

Does MCPBus require external network access?

No. MCPBus runs in 100% off-line local mode by default. It doesn't require external network access unless your APIs from your OpenAPI spec have publicly available endpoints. You can run MCPBus inside your local corporative network no problem.

Does MCPBus transmit any of my data "outside"?

No. The core binary requires no internet connection to parse OpenAPI specs or route local traffic. No telemetry sent or any kind of data retention. We never log, collect, or transmit outside your OpenAPI schemas, API keys, MCP-tools parameters/outputs or prompt payloads.

Does MCPBus require any dependencies to be installed?

No. MCPBus compiles directly into a single, highly optimized, self-contained binary with zero external dependencies. No Python, no NodeJS, no JavaScript VMs. All you need is just MCPBus executable binary.

What MCP version is supported?

MCPBus supports the latest MCP protocol version 2025-11-25.

What MCP-authentication types are supported?

MCPBus does provide to MCP-clients only one authentication type (optional) at the moment: bearer token.

What OpenAPI specification versions are supported?

MCPBus fully supports OpenAPI v2.0 and OpenAPI v3.0 at the moment.

Ok, my APIs are also protected. What OpenAPI security schemes does MCPBus support?

MCPBus does support the following OpenAPI security schemes to call your APIs:

• http type, schemes bearer and basic
• apiKey type

Will MCPBus provide to AI-agents any destructive tools?

No. The destructive operations (DELETE, PUT, PATCH) are blocked by default unless you pass the --openapi-allow-destructive=true CLI-flag.

My OpenAPI spec is big. How can I avoid the LLM context bloat?

If your API translates into more than 10 tools, you will likely deal with LLM context bloat problem. In this situation, most of the context window is taken by MCP-tools definitions and model experiences attention dilution, struggles to decide which tools to call and in which sequence, accuracy of you AI-agent drops significantly but your token-expenses sky-rocket!

With MCPBus, you can build autonomous AI-Agents that have access to hundreds or thousands of different tools, without context bloat.

MCPBus translates all OpenAPI specs into TS/JS SDK and provides just two stable MCP-tools with 806 tokens context footprint:

• search_docs - to search the TS/JS SDK documentation and API reference manual to get functions relevant to user's request.
• execute - to execute a multi-line TypeScript or JavaScript code block inside an isolated high-performance sandbox runtime with full support for modern TypeScript and cutting-edge ECMAScript features including arrow functions, optional chaining, array methods, and modern object destructuring.

My OpenAPI spec has lots of local $ref. What MCPBus does with them?

MCPBUs translates all OpenAPI specs into TS/JS SDK which is available to your agent via MCP-tool search_docs and provides all JSON schemas as d.ts type declarations.

All $ref are resolved into TS/JS type declarations available to your model to apply reasoning.

My OpenAPI spec has local recursive $ref. How MCPBus does deal with them?

All recursive references (or cycled types) are translated into TS/JS type declarations so recursive references will be resolved by named types.

I use MCPBus and love it! How can I get support?

Please check out Support / Contacts