Lab 6 - MCP: the reference
Lab 6 connects the Robot Framework MCP server, has your agent rebuild a Lab 5 test step by step on the live page, and compares the two. Transcript.
The reference
The server's configuration for Claude Code, .mcp.json. The other agents' files are in mcp/.
{
"mcpServers": {
"robotframework": {
"type": "stdio",
"command": "uv",
"args": [
"run",
"--no-sync",
"rf-mcp"
],
"env": {
"PYTHONPATH": "."
}
}
}
}
The stepwise rebuild of WEB-004_AC-3 lost the comparison, so the reference keeps the Lab 5 test and has no stepwise
file.
Why it is a good result
- The server runs from the project's environment at the pinned version, with
PYTHONPATH=., so it finds the shop's settings. - The comparison was decided on the conventions, not on which version is newer, and one test per criterion is left.
What to debrief
What live access changed in the rehearsal, and what it did not:
- Changed: it pinned what the shop happens to show. The stepwise version found the search input by its accessible name, "Search products".
- Not changed: every assumption of the Lab 5 version held: one search box, the results region, one card with its image and price. Neither version can tell whether Enter produced the results, because the shop also searches by itself shortly after typing stops.
- Why the Lab 5 version won: "Search products" comes from the shop, not from the specification, and the conventions allow accessible names only where the specification speaks of a label. The stepwise heading check was case-sensitive, and the stepwise tests covered one of AC-3's two ways of searching.
Compare yours
git fetch upstream solutions
REF=$(git log -1 --format=%H --grep '^lab-06-mcp' upstream/solutions)
git diff "$REF" -- .mcp.json
The lines marked - are the reference's, the lines marked + yours. No upstream remote yet? Add it first.
Your stepwise test is the result to compare: with the conventions and with your Lab 5 test, as the lab's comparison step does.