Add docs for inbound network isolation (#213)

Co-authored-by: Dan Barr <[email protected]>

Author: eleftherias

docs/toolhive/guides-cli/custom-permissions.mdx

@@ -32,7 +32,11 @@ Profiles are defined in JSON format and can include the following properties:
 - `read`: List of paths on your host file system that the MCP server can read
 - `write`: List of file system paths that the MCP server can write to (this also
   implies read access)
-- `network`: Network access rules for outbound connections:
+- `network`: Network access rules for inbound and outbound connections:
+  - `inbound`: Inbound network access rules, which include:
+    - `allow_host`: List of allowed hostnames that can send traffic to the MCP
+      server. If not specified, only the container's own hostname, `localhost`,
+      and `127.0.0.1` are allowed.
   - `outbound`: Outbound network access rules, which include:
     - `insecure_allow_all`: If set to `true`, allows unrestricted outbound
       network access. This isn't recommended for production use.

docs/toolhive/guides-cli/network-isolation.mdx

@@ -242,11 +242,17 @@ without editing a profile file.
 
 ## Accessing other workloads on the same container network
 
+ToolHive allows you to configure both outbound and inbound network access for
+MCP servers. This is commonly needed when your MCP server needs to communicate
+with databases, APIs, or other services that are running on your local host
+during development, or when other containers need to communicate with your MCP
+server.
+
+### Outbound access: MCP server to other workloads
+
 To allow an MCP server to access other workloads on the same network, you need
-to configure network isolation to include the appropriate hostnames and ports.
-This is commonly needed when your MCP server needs to communicate with
-databases, APIs, or other services that are running on your local host during
-development.
+to configure outbound network isolation to include the appropriate hostnames and
+ports.
 
 For example, in Docker environments, you can add `host.docker.internal` to
 access services on the host. `host.docker.internal` is a special hostname
@@ -273,6 +279,45 @@ Run the MCP server with this profile:
 thv run --isolate-network --permission-profile ./internal-access-profile.json <SERVER>
 ```
 
+### Inbound access: Other containers to MCP server
+
+By default, the ingress proxy only allows traffic from the container's own
+hostname, `localhost`, and `127.0.0.1`. If you need to allow other containers or
+workloads to communicate with your MCP server, configure the
+`network.inbound.allow_host` setting in your permission profile.
+
+This is useful when:
+
+- Other containers need to call your MCP server's API
+- You're running multiple services that need to communicate with each other
+- You need to allow traffic from specific internal hostnames or domains
+
+Create a permission profile that allows specific inbound hostnames:
+
+```json title="inbound-access-profile.json"
+{
+  "network": {
+    "inbound": {
+      "allow_host": ["host.docker.internal", "localhost"]
+    }
+  }
+}
+```
+
+Run the MCP server with this profile:
+
+```bash
+thv run --isolate-network --permission-profile ./inbound-access-profile.json <SERVER>
+```
+
+:::info
+
+If no `network.inbound` configuration is specified, the ingress proxy uses the
+default behavior of allowing traffic only from the container's own hostname,
+`localhost`, and `127.0.0.1`.
+
+:::
+
 ## Related information
 
 - [`thv run` command reference](../reference/cli/thv_run.md)