A Model Context Protocol server for Jupiter token swapping on Solana
This MCP server exposes three Solana/Jupiter DeFi tools with visible schemas and basic descriptions. However, there are significant gaps in parameter documentation, output schema clarity, and error handling guidance. Tool names follow verb_noun convention appropriately (jupiter_get_quote, jupiter_build_swap_transaction, jupiter_send_swap_transaction), but parameter descriptions are minimal or missing entirely. The schema for jupiter_send_swap_transaction has NO required parameters despite being a write operation that needs either swapTransaction or serializedTransaction. Output schemas are not documented at all, LLMs cannot predict what data will be returned. Error messages return raw strings without recovery guidance. The server lacks enum constraints for constrained inputs (e.g., swapMode accepts 'string' with no valid values enumerated). Security concern: quoteResponse and serializedTransaction are passed as JSON strings, forcing the tool to parse them, this design invites injection and makes validation responsibility unclear.
Build a swap transaction on Jupiter
Get a quote for swapping tokens on Jupiter
Send a swap transaction on Jupiter
jupiter_send_swap_transaction has ZERO required parameters despite being a destructive write operation. Calling it with no arguments will fail at runtime. Missing required constraint on at least one of swapTransaction or serializedTransaction.
ALL three tools have incomplete parameter descriptions. 'swapMode' is typed 'string' with no enum, valid values, or usage documentation. LLMs will hallucinate values like 'fast', 'slow', 'best_price' without guidance. Similarly, 'excludeDexes' array lacks item description or examples.
No output schemas documented for any tool. LLMs cannot predict what data is returned by jupiter_get_quote (quote structure? price? fees?) or jupiter_build_swap_transaction (transaction object? serialized bytes?). This forces LLMs to guess field names for downstream operations.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 43 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 22 | - | v1 |
Error handling provides no recovery guidance. In src/handlers/jupiter.ts, errors are caught and returned as plain strings: 'Error getting quote: {message}'. No actionable next steps. If API returns 400, LLM has no guidance on what to adjust (slippage? route? amount?).
Tool descriptions do not indicate data dependencies or sequencing. jupiter_build_swap_transaction and jupiter_send_swap_transaction are designed to follow jupiter_get_quote, but the descriptions do not state this. LLMs may attempt to call them out of order or with missing data.
jupiter_send_swap_transaction is marked Risk:WRITE but its description does not warn of irreversible consequences or require confirmation. Per pattern:confirmation-request, destructive operations should support dry-run or confirmation steps. This tool sends real transactions, agents need explicit guidance that this is dangerous.
quoteResponse and serializedTransaction are passed as JSON strings, forcing the tool to parse them. This design is error-prone and makes validation responsibility unclear. If the JSON is malformed, the error is 'Invalid quote response: SyntaxError' with no guidance on format. Consider accepting structured objects or documenting the expected JSON schema inline.
Numeric parameters lack constraints. prioritizationFeeLamports, computeUnitPriceMicroLamports, slippageBps, platformFeeBps, maxRetries, maxAccounts have no min/max bounds. LLMs can pass negative numbers, zero, or absurdly large values without bounds checking.