🛡
+
eng
Integration

Palantir AIP + Engram

Palantir gives Claude the data.
Engram gives Claude the memory.

When an analyst uses palantir-mcp, Claude gets live access to Foundry datasets, ontology objects, and AIP functions. But when the session closes, Claude forgets everything. Tomorrow the analyst starts from zero.

Engram runs alongside palantir-mcp as a second MCP server. Every Foundry session is recorded. Patterns emerge across sessions. The analyst's query trajectory is decoded. Claude enters every new session already knowing where the investigation has been — and where it's heading.

palantir-mcp gives Claude
◈ Live Foundry dataset access in Claude
◈ Ontology object queries
◈ Pipeline + workflow triggers
◈ AIP function execution
Engram adds on top
+ Memory that persists after the session closes
+ Pattern signal: how YOU query Foundry over time
+ Cross-session intelligence — what you asked last week informs today
+ Anomaly detection: when your query pattern breaks
+ Prediction: what dataset you'll need before you ask

Setup in 3 steps

01
Get your Engram API key
Sign in at crossagent.si → Dashboard → API Keys → Generate key. Starts with eng_
Get API key →
02
Add both servers to Claude Desktop
Open ~/Library/Application Support/Claude/claude_desktop_config.json and add both MCP servers:
claude_desktop_config.json
{
  "mcpServers": {
    "palantir": {
      "command": "npx",
      "args": ["@palantir/mcp", "--profile", "your-profile"],
      "env": { "FOUNDRY_URL": "https://your-stack.palantirfoundry.com" }
    },
    "engram": {
      "url": "https://backend-wine-phi-85.vercel.app/mcp",
      "type": "http",
      "headers": {
        "Authorization": "Bearer eng_your_key_here"
      }
    }
  }
}
03
Start a Foundry session
Open Claude Desktop. Run get_context first — Engram injects your pattern signal. Use palantir-mcp tools normally. At the end of your session, run foundry_session to record what you found.

Engram tools for Foundry analysts

foundry_session()

Records what dataset you queried and what insight you derived. Builds your Foundry intelligence profile session by session.

foundry_session(dataset="flight_risk_model", query_intent="evaluating threat vectors in sector 7", insight="anomaly spike on Tuesday correlated with personnel access logs")
pattern_signal()

Returns your cognitive fingerprint across all Foundry sessions — dominant data domain, velocity, and prediction of your next query direction.

pattern_signal() → "Dominant: counterintelligence datasets (34 sessions). Accelerating. Prediction: you are about to pivot to network analysis."
get_context()

Injects your full Foundry session history as a pattern-aware context block. Claude enters every session already knowing your analysis trajectory.

get_context(prompt="reviewing sector 7 anomalies") → PATTERN SIGNAL + relevant past Foundry sessions
recall()

Semantic search across all past Foundry sessions. "What did I find about network anomalies last month?" — answered instantly.

recall(query="network anomaly correlations") → returns your past findings from Foundry, ranked by relevance
What this unlocks

Palantir knows what's in the data.
Engram knows how you think about it.

After 50 Foundry sessions, Engram knows which data domains you gravitate toward, which ontology objects you correlate most often, and when your query pattern is about to shift. That's not Palantir's intelligence about the world — it's intelligence about how you use Palantir. That's the gap. That's what we filled.

Get your Engram API key free →
No credit card. Works with any Palantir AIP stack.