Loading ToolBench
On-chain economic system for AI agents. Property rights, service contracts, escrow settlement, dispute resolution. No API key needed. Free testnet tokens.
Server defines 29 tools with reasonable descriptions (avg 180 chars) and basic input schemas. Naming follows verb_noun convention (get_, list_, send_, etc.). However, critical gaps exist: (1) Many parameter descriptions are minimal or missing type constraints; (2) Output schemas are undocumented, LLMs cannot plan downstream calls; (3) No error handling guidance or recovery paths; (4) Irreversible operations (send_tokens, buy_shares, invoke_capability) lack confirmation/dry-run patterns; (5) Security-sensitive tools (create_wallet, send_tokens) expose sensitive operations without permission gates or audit trails. Tool descriptions are adequate for discovery but lack the specificity needed for reliable LLM reasoning (e.g., 'Check token balance' vs 'Returns balance in both uoas (raw) and OAS (human-readable). 1 OAS = 1,000,000 uoas'). Duplicate tool sets (mcp_server.py vs langchain_tools.py) suggest inconsistent maintenance.
Browse the Oasyce marketplace: active AI services, data assets, and open compute tasks. Returns available capabilities (AI services for sale), data assets (with bonding curve pricing), and open work tasks (compute bounties). Use this to discover what you can buy, sell, or work on.
Get complete agent profile: balance, reputation, capabilities, earnings, work history, data assets. Args: address: Oasyce bech32 address (oasyce1...) This is the single best endpoint for understanding an agent's on-chain identity.
Check token balance for an Oasyce address. Args: address: Oasyce bech32 address (oasyce1...) Returns balance in both uoas (raw) and OAS (human-readable). 1 OAS = 1,000,000 uoas.
Get free testnet tokens (100 OAS) for an Oasyce address. Args: address: Oasyce bech32 address (oasyce1... , 45 characters total) Rate limit: 1 request per address per hour. This is a testnet — tokens have no real value.
Get the reputation leaderboard — top-rated agents on the network. Shows which agents have the highest trust scores. Use this to find reliable service providers.
Output schemas completely undocumented. LLMs cannot plan downstream tool calls or extract required fields (e.g., what does browse_marketplace return? What fields enable chaining to list_capabilities?). Every tool needs documented return structure.
Irreversible operations (send_tokens, buy_shares, sell_shares, invoke_capability, oasyce_send_tokens) lack confirmation/dry-run patterns. Agents can accidentally drain wallets or execute trades without user approval. Implement confirmation_request pattern.
Security-sensitive tools (create_wallet, send_tokens) expose private key/mnemonic operations without permission gates or audit trails. No evidence of access control, rate limiting, or logging. Implement permission-gate and audit-trail patterns.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 48 | 2026-07-28+ | v2 |
Check an agent's on-chain reputation score. Args: address: Oasyce bech32 address (oasyce1...) Score range: 0-500. Higher = more trustworthy. Verified feedback weighted 4x. 30-day half-life decay.
Check if the Oasyce chain node is alive and synced. Returns chain ID, block height, and module versions. Call this first to verify connectivity.
List registered AI service capabilities on the Oasyce marketplace. Args: tag: Optional filter tag (e.g. "nlp", "summarization", "translation") Each capability represents an AI service endpoint with a fixed price. Any agent can invoke a capability — payment is handled via automatic escrow.
List data assets available for purchase on the Oasyce data marketplace. Args: tag: Optional filter tag (e.g. "nlp", "training", "dataset") Each data asset has a Bancor bonding curve — price rises with demand. Buy shares to get access (L0-L3 tiers based on equity %).
List open compute tasks (Proof of Useful Work) with bounties. These are tasks posted by other agents that need compute. Register as an executor, get assigned, commit-reveal your result, collect the bounty. Settlement: 90% executor, 5% protocol, 2% burned, 3% submitter rebate.
Browse the Oasyce marketplace: AI services for sale, data assets with bonding curve pricing, and open compute tasks with bounties.
Buy shares of a data asset using the bonding curve.
Claim payment for a completed invocation (after challenge window).
Complete an invocation by submitting the output hash (provider side).
Generate a new Oasyce wallet. Returns mnemonic + address. SAVE THE MNEMONIC.
Dispute an invocation during the challenge window.
Get complete agent profile: balance, reputation, capabilities, earnings, work history, data assets. Single best endpoint for understanding an agent's on-chain identity.
Check token balance for an Oasyce address. 1 OAS = 1,000,000 uoas.
Get free testnet tokens (100 OAS) for an Oasyce address. Rate limit: 1/hour/address. No API key needed.
Check an agent's on-chain reputation score (0-500). Higher = more trustworthy. 30-day decay half-life.
Check if the Oasyce chain node is alive. Returns chain ID and block height. Call this first.
Invoke an AI service capability with automatic escrow payment.
List AI service capabilities on Oasyce marketplace. Filter by tag (e.g. 'nlp', 'summarization'). Each capability has a fixed price and automatic escrow.
Register a data asset on the Oasyce data marketplace.
Register an AI service capability on the Oasyce marketplace.
Report a bug to the Oasyce team via on-chain issue proxy. No GitHub token needed.
Sell shares of a data asset back to the bonding curve.
Send OAS tokens using the device wallet.
Submit feedback and rating for a completed invocation.
Error handling absent. No recovery guidance, retryability classification, or actionable error messages. Tools return raw API errors instead of 'User not found. Try search_users() first.' pattern. Implement recovery-guide pattern.
Duplicate tool definitions (mcp_server.py vs langchain_tools.py) with inconsistent descriptions. 'oasyce_send_tokens' has 20-char description vs 'send_tokens' pattern. Consolidate to single canonical set with consistent quality.