Cognitive Services MCP Server avatar

Cognitive Services MCP Server

Pricing

from $8.00 / 1,000 novelty gates

Go to Apify Store
Cognitive Services MCP Server

Cognitive Services MCP Server

MCP server giving AI agents a cognitive layer. Three tools: semantic_dedup (detect duplicate knowledge), novelty_gate (should this be stored?), and belief_check (do two ideas agree?). Powered by vector embeddings and Gemini. Pay-per-use.

Pricing

from $8.00 / 1,000 novelty gates

Rating

0.0

(0)

Developer

IDAPIXL

IDAPIXL

Maintained by Community

Actor stats

0

Bookmarked

1

Total users

0

Monthly active users

22 days ago

Last modified

Share

Cognitive Services MCP — Memory Dedup, Novelty & Belief

Give your AI agents a shared cognitive layer. Three tools for managing what gets stored, what gets skipped, and whether two ideas actually agree.


Quick Connect

Claude Desktop (claude_desktop_config.json):

{
"mcpServers": {
"cognitive-services": {
"command": "npx",
"args": ["-y", "@apify/actors-mcp-server", "--actors", "idapixl/cognitive-services-mcp"],
"env": {}
}
}
}

Claude Code (.mcp.json):

{
"mcpServers": {
"cognitive-services": {
"command": "npx",
"args": ["-y", "@apify/actors-mcp-server", "--actors", "idapixl/cognitive-services-mcp"]
}
}
}

Tools

semantic_dedup

Check whether a piece of text is semantically novel or a duplicate of existing knowledge.

Input:

FieldTypeRequiredDescription
textstringYesThe text to check (max 4,000 characters)

Output:

{
"decision": "novel",
"novelty_score": 0.87,
"max_similarity": 0.13,
"nearest_concept": "Transformer architecture"
}
  • decision: novel (safe to store), link (related, add a reference), merge (near-duplicate, skip or merge)
  • novelty_score: 0–1 (1 = fully novel, 0 = exact duplicate)
  • nearest_concept: name of the closest match in memory, if any

Example call:

semantic_dedup("Attention mechanisms allow models to weigh the importance of different tokens when generating each output token.")

novelty_gate

Evaluate whether new information is worth storing and check for contradictions with existing beliefs.

Input:

FieldTypeRequiredDescription
textstringYesThe text to evaluate (max 4,000 characters)

Output:

{
"novelty_score": 0.62,
"recommendation": "store",
"conflicting_belief": null,
"related_concepts": [
{ "name": "Self-attention", "similarity": 0.71 },
{ "name": "BERT", "similarity": 0.58 }
]
}
  • recommendation: store (write it), skip (redundant), merge (consolidate with existing), contradiction_flag (conflicts with a belief — review before storing)
  • conflicting_belief: name of the belief it conflicts with, if any

Example call:

novelty_gate("GPT-4 cannot reason about multi-step logical problems without chain-of-thought prompting.")

belief_check

Check two statements for logical consistency.

Input:

FieldTypeRequiredDescription
statement_astringYesFirst statement (max 4,000 characters)
statement_bstringYesSecond statement (max 4,000 characters)
contextstringNoOptional context to guide the check (max 500 characters)

Output:

{
"verdict": "tension",
"reasoning": "The statements exist in tension — not directly contradictory but pulling in different directions.",
"related_concepts": [
{ "name": "Model scaling", "similarity": 0.83 },
{ "name": "Emergent capabilities", "similarity": 0.77 }
]
}
  • verdict: genuine (direct contradiction), tension (pulling in different directions), complementary (compatible, different angles), unrelated (no meaningful relationship)

Example call:

belief_check(
"Larger models always perform better on reasoning tasks.",
"Smaller fine-tuned models outperform large general models on narrow domains.",
"AI model performance and scale"
)

Pricing

ToolPriceWhat triggers it
semantic_dedup$0.003One deduplication check
novelty_gate$0.008One novelty evaluation (includes contradiction check)
belief_check$0.015One belief consistency evaluation

Pay-per-use. No subscription. No minimum. Charged per successful tool call.


How It Works

This is a proxy actor — it wraps the Idapixl Cortex cognitive service as MCP tools.

Your Agent
│ MCP tool call (JSON-RPC over HTTP)
Apify Standby Actor ←── Express + MCP SDK
│ POST /x402/dedup|novelty|belief
Idapixl Cortex (Cloud Run)
│ Vertex AI Embeddings + Firestore vector search + Gemini LLM
Response → MCP tool result → Your Agent

The cortex service runs Vertex AI text embeddings against a shared memory graph stored in Firestore. Belief checking uses Gemini to evaluate logical consistency.

Stateless design: Every MCP request is isolated. No session state is maintained between calls.


Configuration

InputTypeDefaultDescription
cortexUrlstringhttps://idapixl-cortex-215390428499.us-central1.run.appCortex service base URL
cortexApiKeystring (secret)Optional API key for authenticated access