Stellar Network Intelligence for AI Agents. Payment corridor analysis, tokenized asset checks, issuer trust, anchor compliance, DEX liquidity, stablecoin monitoring, cross-border payment intelligence. Evidence-grade data for institutional payments and tokenized finance on Stellar.
XLMOracle has 5 tools with basic schemas and descriptions, but significant gaps limit production readiness. All tools have descriptions (10-120 chars, mostly adequate) and input schemas with type definitions. However, parameter descriptions are sparse or missing context. Output schemas are not documented, LLMs cannot predict response structure for chaining. No error handling guidance (e.g., 'If account not found, try search_accounts'). Tool names lack clear verb-noun structure (e.g., 'overview' vs 'get_network_overview'). Risk scoring logic is hardcoded and not explained. No pagination support despite returning lists. Baseline: avg tool description ~80 chars (p10=34, p90=392); this server averages ~70 chars, below median. 0 of 5 tools document output schemas.
Stellar account intelligence: balances, trust lines, flags
Stellar anchor intelligence: SEP compliance, fiat on/off ramps
Stellar asset (token) analysis: issuer trust, holders, trading volume
Stellar network overview: XLM price, network stats, ledger info
Cross-border payment corridor intelligence on Stellar
No output schemas documented. LLMs cannot predict response structure (fields, types, nesting) for downstream tool chaining. E.g., handle_account_intel returns risk_score, risk_grade, tokens array, but this is invisible to the agent.
Tool names lack action verbs. 'overview' should be 'get_network_overview'; 'account_intel' should be 'get_account_intelligence'. Verb-noun naming helps LLMs infer intent from the name alone.
Parameter descriptions are minimal or missing context. E.g., 'asset_issuer' says '(optional, uses known issuers if not provided)' but does not explain what 'known issuers' are or how to find them. LLMs cannot reason about optional params without clear guidance.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 44 | <=2025-11-25 | v2 |
No error handling guidance. If account_intel receives an invalid account_id, it returns {'error': 'Account not found: ...'} with no recovery hint. Should suggest 'Try search_accounts() with a partial name' or similar.
Risk scoring logic (risk_grade function) is hardcoded and opaque. Agents cannot understand why a score changed or what actions reduce risk. Scoring rules should be documented in tool descriptions or returned as part of the response.