Agreement not yet measured: the pre-registered rule needs at least three agent pairs; this preview has fewer agents complete so far.
BROWSER · CLAUDE CODE · SONNET 5.516 core tasks · claude-sonnet-5-5 · claude-code 2.1.296
| BAND | SERVER | DETAILS | COMPLETION % |
|---|---|---|---|
| BAND 1 · statistically tied — listed A→Z | |||
| 1 | agent-browser (Vercel Labs)29 tools · run log · repo | coverage —% · reliability —% · extended 100% median call 0.1s SHRED |
100100–100 |
| 1 | @agent-infra/mcp-server-browser (ByteDance)21 tools · run log · repo | coverage —% · reliability —% · extended 100% median call 0.0s SHRED |
100100–100 |
| 1 | browser-use (MCP mode)16 tools · run log · repo | coverage —% · reliability —% · extended 100% median call 0.6s SHRED |
9481–100 |
| 1 | Chrome DevTools MCP (Google)30 tools · run log · repo | coverage —% · reliability —% · extended 100% median call 0.2s SHRED |
100100–100 |
| 1 | Playwright MCP (ExecuteAutomation)33 tools · run log · repo | coverage —% · reliability —% · extended 100% median call 0.1s SHRED |
9481–100 |
| 1 | Lightpanda32 tools · run log · repo | coverage —% · reliability —% · extended 100% median call 0.0s SHRED |
9176–100 |
| 1 | Playwright MCP (Microsoft)25 tools · run log · repo | coverage —% · reliability —% · extended 100% median call 0.2s SHRED |
100100–100 |
| REFERENCE ROWS · archived/deprecated servers, scored but not ranked | |||
| REF | Puppeteer (archived official)7 tools · run log · repo | coverage —% · reliability —% · extended 100% median call 0.0s SHRED |
100100–100 |
Completion = share of 16 core jobs finished, averaged per task over 2–5 attempts (95% interval below the score). Bands group servers whose rank intervals overlap. Coverage = share of capabilities the server advertises; reliability = pass rate on the jobs it claims to support.
BROWSER · CODEX CLI · GPT-5.6 TERRA16 core tasks · gpt-5.6-terra · codex-cli 0.162.0 · LOW-SAMPLE
| BAND | SERVER | DETAILS | COMPLETION % |
|---|---|---|---|
| NOT RANKED · incomplete coverage this sweep | |||
| — | Playwright MCP (Microsoft)25 tools · run log · repo | coverage —% · reliability —% median call 0.1s |
51 |
| — | Chrome DevTools MCP (Google)30 tools · run log · repo | coverage —% · reliability —% median call 0.2s |
56 |
| — | agent-browser (Vercel Labs)29 tools · run log · repo · ⚑ >5% excluded attempts | coverage —% · reliability —% median call 0.0s |
44 |
| — | Playwright MCP (ExecuteAutomation)33 tools · run log · repo · ⚑ >5% excluded attempts | coverage —% · reliability —% median call 0.0s |
49 |
| — | browser-use (MCP mode)16 tools · run log · repo | coverage —% · reliability —% median call 0.6s |
45 |
| — | @agent-infra/mcp-server-browser (ByteDance)21 tools · run log · repo | coverage —% · reliability —% median call 0.0s |
41 |
| — | Lightpanda32 tools · run log · repo | coverage —% · reliability —% median call 0.0s |
45 |
| — | Puppeteer (archived official)7 tools · run log · repo | coverage —% · reliability —% median call 0.0s |
44 |
Completion = share of 16 core jobs finished, averaged per task over 2–5 attempts (95% interval below the score). Bands group servers whose rank intervals overlap. Coverage = share of capabilities the server advertises; reliability = pass rate on the jobs it claims to support.