Loading ToolBench

Reading the current corpus…

agentshare-mcp MCP server · B · ToolBench

agentshare-mcp

14 MCP tools on agent-paid rails (dual-auth API key or x402): commerce, request_capability, Market experiment (market_browse / market_list), plus secondary Solana/Meteora DeFi demos. Polymarket is not on the public MCP catalog. Streamable HTTP at agentshare.dev/mcp.

2026-07-28+STREAMABLE_HTTP7 tools

73/100

Definition Quality73

Protocol Readiness72

SupportabilityNot measured-

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence · evidence: stateless requests, Streamable HTTP

Current-spec patterns detected

Summary

AgentShare exposes 7 tools (of 14 declared) with mixed quality. Strong points: most tools have detailed descriptions (100-250 chars, aligned with baseline), clear action verbs in names (search_, best_offer, service_meta), and input schemas with types and ranges. Weak points: only 3 tools have visible, complete input schemas; 4 tools inferred from mcp.json metadata with no schema validation visible in source code; parameter descriptions vary widely in specificity; output schemas are documented only in comments, not returned as structured definitions; error handling strategies are present in mcp_tool_format.py but not linked to individual tools; tool annotations present (toolAnnotations=true declared) but not visible in source for most tools.

Tools (7)

best_offerread onlyauthsource verified82/100

Return the single best current offer for a product intent: typically lowest price among in-stock listings the API trusts. Use when the user asks where to buy something cheapest, 'best deal', or one clear recommendation. For side‑by‑side comparison of many listings, prefer search_products.

best_offer_under_budgetread onlyauthsource verified77/100

Find the best offer for a product query with a maximum price ceiling. Use when the user gives a budget, 'under $X', 'below …', or 'no more than …'. Pass max_price in the same numeric unit the deployed API expects for that field (see API docs). Do not use for open-ended comparison without a cap—use search_products or best_offer.

dex_overviewread onlyauthsource verified67/100

Macro DEX protocol rankings by 24h volume (DefiLlama aggregated). Use chain=solana for AgentShare's primary focus; other chains may be available upstream.

meteora_briefread onlyauthsource verified71/100

Ranked Meteora DLMM pool brief for Solana (momentum, fee velocity, SAFE/CAUTION/AVOID). Primary AgentShare DeFi endpoint for autonomous agents. Live data from agentshare.dev.

search_productsread onlyauthsource verified83/100

Search connected marketplaces and return structured offers (prices, sources, freshness). Use when the user wants to compare options, browse multiple listings, or explore a product category or model—not when they only need one definitive 'cheapest' pick (use best_offer). Works for robotics parts, robot/drone batteries, and AI hardware — see /coverage for focus areas. Accepts free-text queries in any language.

Issues (7)

CRITICALtool ↗

4 tools (dex_top_movers, meteora_pool_detail, product_detail, commerce_quote, request_capability, market_browse, market_list) declared in mcp.json but NO visible schema or source code provided. These tools are inferred from metadata only, cannot validate schemas, parameters, or error handling.

dex_top_moversmeteora_pool_detailproduct_detailcommerce_quoterequest_capabilitymarket_browsemarket_list

HIGHtool ↗

Output schemas documented only in comments (mcp_tool_format.py, build_summary function) as template patterns, not as machine-readable JSON Schema declarations. LLMs need formal schema to parse responses reliably.

search_productsbest_offerbest_offer_under_budget

MEDIUMconstrained-input ↗

Parameter descriptions vary in completeness. 'limit' parameters described well (e.g. 'Maximum rows to return (higher = broader scan, more tokens)'), but 'window' in meteora_brief lacks guidance on valid values and impact ('5m', '1h', '6h', '24h' are examples, not constraints).

Recommendations

Publish full tool definitions (input + output schemas as JSON Schema) for all 14 declared tools. Currently only 7 visible in source; 7 inferred from metadata only.
Add formal enums to parameters accepting constrained values. E.g. 'window' in meteora_brief should be enum: ['5m', '1h', '6h', '24h'] instead of free-text description.
Link error codes (INVALID_QUERY, NOT_FOUND, RATE_LIMITED, STALE_DATA, UPSTREAM_ERROR) defined in mcp_tool_format.py to individual tool descriptions with recovery guidance. Example: 'If RATE_LIMITED, retry after 60 seconds; contact support@agentshare.dev if recurring.'
Document pagination strategy for high-limit queries (search_products limit=50, dex_overview limit=50). Add cursor, offset, or page parameters and total_count to responses.
Add permission scope declarations to API key docs. Map each tool to required scopes (e.g. search_products requires 'read:commerce', meteora_brief requires 'read:defi'). Enable least-privilege agent configurations.
Remove external documentation references from tool descriptions. Inline all constraints, examples, and valid enums. E.g. instead of 'see /coverage for focus areas', state: 'Covers robotics parts, robot/drone batteries, and AI hardware. Other categories not supported.'
Formalize output schemas as JSON Schema objects returned in tool documentation. Document fields, types, required properties, and sample responses for each tool.
Add dry-run or confirmation step to any tools that trigger external payments or trades (if applicable). Agents must not execute financial transactions without explicit user consent.

Score history

Overall score trend

First recorded score · v2 rubric

73/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	B	73	2026-07-28+	v2