Wanaku

What wanaku-ai/wanaku does

The wanaku-ai/wanaku MCP server acts as an intermediary between AI agents and the systems those agents need to affect. Agents submit tool calls to Wanaku, while the proxy performs the corresponding backend work. This separation keeps direct backend access out of the agent and provides a central location for policy, identity, data controls, and auditing.

Integration developers can publish Apache Camel routes as agent-facing tools. Wanaku can also register remote MCP servers as forwards, discover their tools, and make those tools available through its local catalog. Tools and resources can be organized into namespaces for teams, tenants, or environments.

The project includes a management API for tools, resources, prompts, forwards, and namespaces, plus a web-based administration interface. Its architecture supports plugins through feature crates and a composable filter pipeline.

How it works

Clients connect to the MCP endpoint at http://localhost:8081/mcp. Namespaced access uses a path such as /{namespace}/mcp. The management API runs separately on port 8080 under /api/v1/..., and the admin interface is available at /admin/.

A remote MCP server is added as a forward with a name and address. After a refresh operation, Wanaku discovers the upstream tools and exposes them in its local catalog. The client does not need to know that those tools are forwarded. The proxy can intercept tool calls, agent-to-agent messages, and inference traffic, depending on the configured deployment and controls.

Policy enforcement can use LLM-powered evaluators and WebAssembly action scripts to classify, filter, or block tool calls. Authentication and authorization are provided through an external oauth2-proxy and Keycloak arrangement rather than being configured solely inside the server.

Setup and configuration

The wanaku-ai/wanaku MCP server can be installed on Linux or macOS with the repository's shell installer, which places wanaku-server in $HOME/bin. The installation directory can be changed with WANAKU_INSTALL_DIR. A container image is also published as quay.io/wanaku/wanaku-server, with MCP and management ports exposed separately.

A wanaku.yaml file can preload forward definitions. For example, a forward specifies a local name and the URL of an upstream MCP endpoint. Without a configuration file, the server starts with an empty registry that can be populated through the management API.

Building from source requires Rust 1.96 or later and Yarn. Running cargo build also builds the admin UI when its distribution directory is absent. The documented source workflow then starts the service with cargo run.

For protected deployments, oauth2-proxy instances are placed in front of the MCP and management ports. The MCP proxy accepts authenticated users, while the management proxy requires an administrator role. Wanaku can publish RFC 9728 protected-resource metadata, using WANAKU_AUTH_ISSUER to identify the Keycloak authorization server.

Tools and capabilities

The wanaku-ai/wanaku MCP server provides the following supported capabilities:

• Expose MCP tools through a local or namespaced endpoint.
• Register remote MCP servers as forwards.
• Discover tools from registered upstream MCP servers.
• Publish Apache Camel routes as tools.
• Manage tools, resources, prompts, forwards, and namespaces through the REST API.
• Apply policy filters, WebAssembly action scripts, and LLM-based evaluators.
• Manage configuration through the web admin dashboard.
• Isolate tools and resources by namespace.

Limitations and notes

Wanaku is a routing and governance layer, not a catalog of backend integrations by itself. The README describes Apache Camel routes, upstream MCP servers, and separate capability projects as the mechanisms for adding actions.

Authentication is documented as an external deployment using oauth2-proxy and Keycloak. The repository does not identify a built-in list of compatible desktop clients, nor does it describe a hosted SaaS endpoint. The project is marked as an early-access build, and the supplied quick-start instructions target Linux, macOS, containers, or source builds.

Getting started with this wanaku-ai/wanaku MCP server

Always refer to the official documentation for the most accurate and up-to-date information.