crosscheck-mcp 0.2.2 → 0.2.3

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/pricing.json CHANGED
@@ -3,7 +3,7 @@
3
3
  "currency": "USD",
4
4
  "unit": "per_1k_tokens",
5
5
  "updated_at": "2026-06-07",
6
- "notes": "Prices reflect public list pricing per provider. cached_per_1k applies to prompt tokens served from prompt-cache (Anthropic) or input-cache (OpenAI) when supported. Update when providers change rates. Missing models fall back to cost=0 with estimated=true. 2026-06-07: removed decommissioned gemini-2.0-flash / gemini-2.0-flash-lite (Google returns 404) and replaced the low-tier lite model with gemini-2.5-flash-lite. 2026-06-08: bumped the default Anthropic model to claude-opus-4-8. Kept gemini-2.5-pro as the Gemini default (gemini-3.1-pro-preview measured ~2.2x the cost at $2/$12 per M — output-token dominated by ~1.5k thinking tokens — so not worth it for a 1-of-4 panel voice). 2026-06-09: corrected rates verified against live provider pricing pages — gemini-2.5-pro output $5->$10/M; gemini-2.5-flash $0.15/$0.60 -> $0.30/$2.50; gpt-5 $5/$15 -> $1.25/$10 (cached $0.125); grok-4-latest input $5->$3/M (legacy grok-4 tier). Anthropic opus/sonnet/haiku rates confirmed unchanged. Older/unused models (gpt-4o, o1, o3, grok-3) not all re-verified. 2026-07-02: added claude-fable-5 as the optional reasoning-upgrade model (used only for the lead reasoning seat when the operator bumps a heavy task). 2026-07-04: set claude-fable-5 to its real pay-as-you-go API price effective 2026-07-07 — $10/M input ($0.01/1k), $50/M output ($0.05/1k), cached at 10% of input ($0.001/1k). Dropped the estimated flag. Note: Fable 5 is cheaper than opus-4-8 on both input and output. 2026-07-09: Google retired gemini-2.5-pro AND gemini-3-pro-preview (both now 404 'no longer available'). Moved the Gemini default to gemini-3.1-pro-preview ($2/M in, $12/M out, cache $0.20/M — the newest working pro; note it's a preview and may need another bump) and added gemini-3.5-flash ($1.50/M in, $9/M out, cache $0.15/M — GA) to the low/cheap tier. Removed the dead gemini-2.5-pro rate. IMPORTANT: gemini-3.x are heavy thinking models — thoughtsTokenCount (billed as output) is now folded into completion_tokens in the adapter, so cost reflects reasoning tokens."
6
+ "notes": "Prices reflect public list pricing per provider. cached_per_1k applies to prompt tokens served from prompt-cache (Anthropic) or input-cache (OpenAI) when supported. Update when providers change rates. Missing models fall back to cost=0 with estimated=true. 2026-06-07: removed decommissioned gemini-2.0-flash / gemini-2.0-flash-lite (Google returns 404) and replaced the low-tier lite model with gemini-2.5-flash-lite. 2026-06-08: bumped the default Anthropic model to claude-opus-4-8. Kept gemini-2.5-pro as the Gemini default (gemini-3.1-pro-preview measured ~2.2x the cost at $2/$12 per M — output-token dominated by ~1.5k thinking tokens — so not worth it for a 1-of-4 panel voice). 2026-06-09: corrected rates verified against live provider pricing pages — gemini-2.5-pro output $5->$10/M; gemini-2.5-flash $0.15/$0.60 -> $0.30/$2.50; gpt-5 $5/$15 -> $1.25/$10 (cached $0.125); grok-4-latest input $5->$3/M (legacy grok-4 tier). Anthropic opus/sonnet/haiku rates confirmed unchanged. Older/unused models (gpt-4o, o1, o3, grok-3) not all re-verified. 2026-07-02: added claude-fable-5 as the optional reasoning-upgrade model (used only for the lead reasoning seat when the operator bumps a heavy task). 2026-07-04: set claude-fable-5 to its real pay-as-you-go API price effective 2026-07-07 — $10/M input ($0.01/1k), $50/M output ($0.05/1k), cached at 10% of input ($0.001/1k). Dropped the estimated flag. Note: Fable 5 is cheaper than opus-4-8 on both input and output. 2026-07-09: Google retired gemini-2.5-pro AND gemini-3-pro-preview (both now 404 'no longer available'). Moved the Gemini default to gemini-3.1-pro-preview ($2/M in, $12/M out, cache $0.20/M — the newest working pro; note it's a preview and may need another bump) and added gemini-3.5-flash ($1.50/M in, $9/M out, cache $0.15/M — GA) to the low/cheap tier. Removed the dead gemini-2.5-pro rate. IMPORTANT: gemini-3.x are heavy thinking models — thoughtsTokenCount (billed as output) is now folded into completion_tokens in the adapter, so cost reflects reasoning tokens. 2026-07-13: bumped the default OpenAI model to gpt-5.5 ($5/M in, $30/M out, cached $0.50/M — verified live). Note gpt-5.5 is markedly pricier than gpt-5 ($1.25/$10): ~4x input, 3x output. Kept gpt-5 in the table for cost lookups on pinned installs."
7
7
  },
8
8
  "anthropic": {
9
9
  "claude-opus-4-5": { "prompt_per_1k": 0.015, "completion_per_1k": 0.075, "cached_per_1k": 0.0015 },
@@ -17,6 +17,7 @@
17
17
  "claude-3-5-haiku": { "prompt_per_1k": 0.0008, "completion_per_1k": 0.004, "cached_per_1k": 0.00008 }
18
18
  },
19
19
  "openai": {
20
+ "gpt-5.5": { "prompt_per_1k": 0.005, "completion_per_1k": 0.03, "cached_per_1k": 0.0005 },
20
21
  "gpt-5": { "prompt_per_1k": 0.00125,"completion_per_1k": 0.01, "cached_per_1k": 0.000125 },
21
22
  "gpt-5-mini": { "prompt_per_1k": 0.0005, "completion_per_1k": 0.0015, "cached_per_1k": 0.00025 },
22
23
  "gpt-4o": { "prompt_per_1k": 0.0025, "completion_per_1k": 0.01, "cached_per_1k": 0.00125 },
@@ -85,6 +86,7 @@
85
86
  { "provider": "anthropic", "model": "claude-opus-4-5" },
86
87
  { "provider": "anthropic", "model": "claude-opus-4-7" },
87
88
  { "provider": "anthropic", "model": "claude-fable-5" },
89
+ { "provider": "openai", "model": "gpt-5.5" },
88
90
  { "provider": "openai", "model": "gpt-5" },
89
91
  { "provider": "openai", "model": "o1" },
90
92
  { "provider": "openai", "model": "o3" },
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "crosscheck-mcp",
3
- "version": "0.2.2",
3
+ "version": "0.2.3",
4
4
  "description": "Multi-LLM MCP server: confer / debate / coordinate / audit / orchestrate across Anthropic + OpenAI + xAI + Gemini + Mistral + Groq + DeepSeek, with scoreboard-driven router, canary-leak detection, sandboxed shell verifiers, and a tier-aware cheap-mode picker. TypeScript implementation; no Python at runtime.",
5
5
  "keywords": [
6
6
  "mcp",