diff --git a/docs/mkdocs.yml b/docs/mkdocs.yml
index e88a87c7..e2440d2e 100644
--- a/docs/mkdocs.yml
+++ b/docs/mkdocs.yml
@@ -83,6 +83,7 @@ nav:
     - Client Libraries: reference/client-libraries.md
     - Credentials: reference/creds.md
     - reference/node-domain.md
+    - Testserver: reference/testserver.md
     - Certificates: reference/certs.md
     - Security: reference/security.md
     - LSP Integration: reference/lsp.md
diff --git a/docs/src/reference/testserver.md b/docs/src/reference/testserver.md
new file mode 100644
index 00000000..49fbcc45
--- /dev/null
+++ b/docs/src/reference/testserver.md
@@ -0,0 +1,111 @@
+# The `gl-testserver`
+
+The `gl-testserver` is a standalone version, of the `gl-testing`
+framework, allowing you to test against a mock Greenlight server,
+independently of your programming language and development
+environment.
+
+The goal of the `gl-testing` package is to enable developers to test
+locally against a mock Greenlight server. This has a number of
+advantages:
+
+ - **Speed**: by not involving the network, you can test without
+   latency slowing you down. The tests also run on a `regtest`
+   network, allowing you to generate blocks and confirm transactions
+   without the need to wait.
+ - **Cost**: the `prod` network is not free, and since tests may
+   consume arbitrary resources, which are then not cleaned up (see
+   next point), repeatedly running them could incur costs. We see this
+   as a bad incentive to minimize testing during development, and
+   `gl-testing` allows you to use only local resources that can then
+   also be reclaimed, making testing free, and hopefully encouraging
+   to test more.
+ - **Reproducibility**: The `prod` network does not allow cleaning up
+   test resources, since there might be an actual user using
+   them. This means that test artifacts persist between runs,
+   resulting in a non-reproducible environment for
+   testing. `gl-testing` running locally allows cleaning up resources,
+   thus enabling reproducible tests.
+   
+However, the downside of `gl-testing` is that is coupled with `python`
+as programming language and `pytest` as test runner. This is where
+`gl-testserver` comes in: by bundling all the fixtures and startup
+logic into a standalone binary we can pull up an instance in a matter
+of seconds, test and develop against it, and then tear it down at the
+end of our session.
+
+## How to use `gl-testserver`
+
+It's probably easiest to use `uv` to run the script from the source
+tree. Please see the [`uv` installation instructions][uv/install] for
+how to get started installing `uv` then come back here.
+
+Executing `uv run gltestserver` is the entrypoint for the tool:
+
+```bash
+gltestserver
+Usage: gltestserver [OPTIONS] COMMAND [ARGS]...
+
+Options:
+  --help  Show this message and exit.
+
+Commands:
+  run  Start a gl-testserver instance to test against.
+```
+
+Currently there is only the `run` subcommand which will start the
+testserver, binding the scheduler GRPC interface, the `bitcoind` RPC
+interface, and the GRPC-web proxy to random ports on `localhost`.
+
+
+```bash
+gltestserver run --help
+Usage: gltestserver run [OPTIONS]
+
+  Start a gl-testserver instance to test against.
+
+Options:
+  --directory PATH  Set the top-level directory for the testserver. This can
+                    be used to run multiple instances isolated from each
+                    other, by giving each isntance a different top-level
+                    directory. Defaults to '/tmp/'
+  --help            Show this message and exit
+```
+
+In order to identify the ports for the current instance you can either
+see the end of the output of the command, which contains
+pretty-printed key-value pairs, or load the `metadata.json` file
+containing the port references to use from the `gl-testserver`
+sub-directory (`/tmp/gl-testserver` if you haven't specified the
+`--directory` option).
+
+!!! note "Running multiple tests in parallel"
+    As the help text above already points out it is possible to run as many
+	instances of the testserver concurrently as you want, by specifying
+	separate `--directory` options in each call.
+	
+	This is particularly useful if you want to run multiple tests in parallel
+	to speed up the test runs. It is also the core reason why ports are
+	randomized rather than using fixed ports per interface, as concurrent
+	instances would conflict, and the isolation between tests could be
+	compromised.
+	
+Once started you will see the following lines printed:
+
+```
+Writing testserver metadata to /tmp/gl-testserver/metadata.json
+{
+  'scheduler_grpc_uri': 'https://localhost:38209',
+  'grpc_web_proxy_uri': 'http://localhost:35911',
+  'bitcoind_rpc_uri': 'http://rpcuser:rpcpass@localhost:44135'
+}
+Server is up and running with the above config values. To stop press Ctrl-C.
+```
+
+At this point you can use the URIs that are printed to interact with
+the services, or use `Ctrl-C` to stop the server. When running in a
+test environment you can send `SIGTERM` to the process and it'll also
+shut down gracefully, cleaning up the processes, but leaving the data
+created during the test in the directory.
+
+[uv/install]: https://docs.astral.sh/uv/getting-started/installation/