PDX-0: chore(scripts): add token-measure-vs-playwright.cjs

[mrdailey99]

Summary

• Promotes scripts/token-measure-vs-playwright.cjs from the PDX-482 worktree to develop so it lives as a permanent, discoverable artifact.
• The script is referenced as the reproduction recipe in the methodology appendix of the published Provar MCP vs. Playwright + AI Coding Agents comparison deck. Without this PR, the methodology slide points at a path that does not exist on develop — undermining the deck's credibility.

Why a separate PR

This is a docs / marketing-support tool, not part of PDX-482's tool-description hardening. Bundling it would muddle PR scope. PDX-0 fits because there is no observable user or system behaviour change — this is purely a scripts/ addition.

What the script does

1. Spawns Provar MCP locally (via bin/mcp-start.js) and Playwright MCP via npx -y @playwright/mcp.
2. Sends the standard MCP initialize → tools/list JSON-RPC pair to each.
3. Serializes the tools[] array — exactly what a real MCP client sends to its LLM as the tool catalog — and counts characters → tokens at chars / 4.
4. Runs Provar MCP in three configurations: STANDARD, COMPACT (PROVAR_MCP_SCHEMA_MODE=compact), AUTHORING (compact + PROVAR_MCP_TOOLS=authoring,inspect,connection,validation).
5. For Playwright MCP, additionally issues browser_navigate + browser_snapshot against example.com to capture a representative per-interaction baseline.

Sample output

═════════════════════════════════════════════════════════════════════════
Scenario                                              Tools  ~Tokens
═════════════════════════════════════════════════════════════════════════
Provar MCP — STANDARD                                    41   18346
Provar MCP — COMPACT                                     41   11750
Provar MCP — AUTHORING                                   21    7897
Playwright MCP — DEFAULT (out-of-the-box)                23    4271
═════════════════════════════════════════════════════════════════════════

Test plan

• yarn install, yarn compile, yarn lint all clean against the chore worktree
• Script runs successfully against both servers and produces the table above
• No source-tree changes; no behavioural impact; no test suite changes required (script is invocation-time only, never imported)
• Reviewer can reproduce by running node scripts/token-measure-vs-playwright.cjs from the repo root after yarn install && yarn compile

Jira

PDX-0 — chore (no Jira ticket). See the comparison deck methodology slide for the customer-facing context this script supports.

🤖 Generated with Claude Code

[github-actions]

Quality Orchestrator

🟢 LOW · 0 / 100 · All changed files have mapped tests.

---

No test files mapped.

---

_{⚡ quality-orchestrator  ·  /qo stub <file>  ·  qo analyze-local}

[Copilot]

Pull request overview

Adds a one-off measurement script that compares MCP tool-catalog token sizes between Provar MCP (in three configurations) and Playwright MCP, used as the reproduction recipe for a marketing comparison deck.

Changes:

• New scripts/token-measure-vs-playwright.cjs spawns each MCP server over stdio, drives initialize → tools/list, and reports catalog chars/tokens.
• For Playwright MCP, additionally measures a browser_navigate + browser_snapshot round trip against example.com as a per-interaction baseline.
• Prints a formatted comparison table, Playwright/Provar ratios, and the top-5 heaviest tools per server.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.