test: Add CLI integration test coverage for MCP

docs/use/execmd/index.md

@@ -57,7 +57,7 @@ The tool does the following:
 
 * The line is not directly `exec`, but passed as-is to `bash -c`
 * The working directory is set to the processed Markdown file's directory (use e.g. ` ```bash cd ..` to change it)
-* Command must exit with `0`, otherwise the tool aborts and returns that code (unless ` ```bash $?`)
+* Command must exit with `0`, otherwise the tool aborts and returns that code (unless ` ```bash $?` or `$%` is used)
 * Command is killed (times out) after 7s if it "hangs" (useful on CI)
 * `STDOUT` & `STDERR` are both captured, and interspersed in MD
 * `STDIN` is closed (but you can use `< ...`)

docs/use/mcp/index.md

@@ -24,23 +24,19 @@ This is similar to the [MCP Inspector](https://github.com/modelcontextprotocol/i
 
 ## Call MCP Tool
 
-<!-- TODO Re-add the 'bash cd ../.././..' header AFTER https://github.com/enola-dev/enola/issues/1862 is fixed... -->
-
-```sh
+```bash cd ../.././..
 $ ./enola -v mcp call-tool modelcontextprotocol/everything echo '{"message":"hi"}'
 ...
 ```
 
 ## List MCP Tools
 
-<!-- TODO Re-add the 'bash $? cd ../.././..' header AFTER https://github.com/enola-dev/enola/issues/1862 is fixed... -->
-
-```sh
+```bash $% cd ../.././..
 $ ./enola mcp list-tools --help
 ...
 ```
 
-<!-- TODO This doesn't work?!
+<!-- TODO Why does this not work?! Is it just because secrets are missing on CI? See also test-cli.bash ...
 
 ```bash cd ../.././..
 $ ./enola -vv mcp list-tools

test-cli.bash

@@ -20,6 +20,10 @@ set -euox pipefail
 # This script tests Enola CLI invocations.
 # See also EnolaCLITest
 
-# ...
+# Integration test the MCP related stuff; initially added to catch https://github.com/enola-dev/enola/issues/1862 regressions:
+# TODO FIXME ./enola -vv mcp list-tools
+./enola -vv mcp call-tool modelcontextprotocol/everything echo '{"message":"hi"}'
+# TODO The ai sub-command currently doesn't actually return 1 instead of 0 on errors, so this is not really not testable yet, but still useful:
+./enola -vv ai --agents=test/agents/everything.agent.yaml --prompt "Print environment variables to debug MCP"
 
 # PS: Update tika.md with anything (of interest) added here

test.bash

@@ -79,9 +79,10 @@ else # On CI
   # https://github.com/enola-dev/enola/issues/1780
   "$BZL" query //... | xargs "$BZL" test --experimental_ui_max_stdouterr_bytes=-1
   cat ~/.bazel/execroot/_main/bazel-out/k8-fastbuild/testlogs/java/dev/enola/cli/tests/test.log || true
-fi
 
-./test-cli.bash
+  # Run Enola CLI integration tests!
+  ./test-cli.bash
+fi
 
 # The following makes sure that this test.bash will run as a pre-commit hook.
 # NB: We DO NOT want to "pre-commit install" because that won't run Bazel!

test/agents/everything.agent.yaml

@@ -1,4 +1,4 @@
 $schema: https://enola.dev/ai/agent
-model: google://?model=gemini-2.5-flash-lite
+model: echo:/
 tools:
   - modelcontextprotocol/everything

tools/test-ci/test.bash

@@ -28,6 +28,8 @@ nix run .#test
 # Skippping test, because Nix did already just run this...
 # ./test.bash
 
+# NB: The //test-cli.bash script is run from within test.bash (if we're on CI)
+
 # Ensure non-regression on exploratory possible future entirely non-Bazel based build infrastructure idea...
 tools/javac/build.bash