@agentwonderland/mcp 0.1.27 → 0.1.28

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (3) hide show
  1. package/dist/index.js +29 -27
  2. package/package.json +1 -1
  3. package/src/index.ts +29 -27
package/dist/index.js CHANGED
@@ -24,40 +24,42 @@ export async function startMcpServer() {
24
24
  version: "0.2.0",
25
25
  }, {
26
26
  instructions: [
27
- "Agent Wonderland — AI agent marketplace. Discover, run, and rate agents.",
27
+ "Agent Wonderland — on-demand AI agent marketplace with autonomous per-call payment. Discover, run, rate.",
28
28
  "",
29
- "FILE HANDLING:",
30
- "Pass local file paths directly as input values (e.g. image: \"/path/to/photo.jpg\").",
31
- "The MCP auto-uploads to cloud storage and replaces with a URL. Never base64-encode.",
29
+ "PURPOSE:",
30
+ "You are an AI agent calling this MCP to extend your capabilities. Instead of integrating N APIs with N keys,",
31
+ "you pay specialist agents per call in USDC. This works headlessly in automations (n8n, cron) and interactively",
32
+ "(Claude Code, Desktop). Favor one-shot `solve()` for autonomous use; prefer `search_agents()` + `run_agent()`",
33
+ "when the caller wants to pick a specific agent.",
32
34
  "",
33
35
  "WORKFLOW:",
34
- "1. search_agents() or solve() — find agents for the task",
35
- "2. get_agent() — check required input fields before running",
36
- "3. run_agent() or solve() with ALL required fields",
37
- "3a. If the agent offers discounted credit packs, use buy_agent_credit_pack() and list_agent_credit_packs() when helpful",
38
- "4. If no payment method is set up, run_agent returns a setup page to connect a credit card — present it to the user",
39
- "5. Ask user to rate or tip after a successful run",
36
+ "1. solve(intent, input, budget) — one call: find best agent + pay + run. Use when the task is clear.",
37
+ " OR: search_agents() → get_agent() to inspect schema → run_agent() with required fields.",
38
+ "2. If the agent returns status 'processing', poll with get_job(). Async runs resolve automatically.",
39
+ "3. After a successful run, rate_agent() and optionally tip_agent() if the result was useful.",
40
+ "4. Use list_jobs() to recover state across sessions (it checks every configured wallet).",
40
41
  "",
41
- "PAYMENT:",
42
- "- Crypto wallets (Tempo USDC, Base USDC, Solana USDC) are the core supported rails.",
43
- "- Saved cards can also be used, but card-backed MPP depends on Stripe Shared Payment Token availability in the current environment.",
44
- "- Card and crypto are SEPARATE payment methods. Card charges a credit card; crypto sends USDC on-chain.",
45
- "- When a card is configured, it becomes the default payment method. Use wallet_status to confirm whether `Card MPP: ready` before relying on it.",
46
- "- run_agent() and solve() auto-detect the default compatible payment method when pay_with is omitted.",
47
- "- Include pay_with explicitly when you need a specific rail or want deterministic control over the payment method used.",
48
- "- Use wallet_status to see the exact payment methods the user has configured before calling a paid tool.",
49
- "- Payment is automatic once configured. Users are never charged for failed runs.",
50
- "- Do NOT ask the user to set up payment before they try to run an agent. Let them explore freely.",
51
- "- If a specific payment method fails, report the error clearly. Do NOT silently fall back to a different method.",
52
- "- When payment fails, suggest alternatives using wallet_status.",
42
+ "PAYMENT (crypto-only right now):",
43
+ "- Supported rails: Tempo USDC, Base USDC, Solana USDC. Card is temporarily disabled pending Stripe SPT approval.",
44
+ "- Tempo and Base share one EVM wallet key. Solana uses a separate ed25519 key. One OWS wallet can manage both.",
45
+ "- If pay_with is omitted, the MCP auto-selects a compatible configured rail. Pass pay_with explicitly",
46
+ " (tempo | base | solana | wallet-id) for deterministic behavior.",
47
+ "- Payment is automatic: on a 402 challenge the MCP signs on-chain, submits, then retries. Failed runs are refunded.",
48
+ "- If a specific rail fails, surface the real reason — do NOT silently retry with a different method.",
49
+ "- Headless/automation: set wallet_set_policy() to cap max_per_tx and max_per_day so a runaway loop can't drain funds.",
53
50
  "",
54
- "MANAGING PAYMENT METHODS:",
55
- "- To remove a card: wallet_setup({ action: \"remove-card\" })",
56
- "- To remove or modify a crypto wallet: NEVER do this programmatically. Direct users to edit their config files manually (~/.agentwonderland/config.json or ~/.ows/). We do not handle private key deletion or modification to avoid any risk of key loss.",
51
+ "FILE HANDLING:",
52
+ "Pass local file paths directly as input values (e.g. image: \"/path/to/photo.jpg\"). The MCP auto-uploads to",
53
+ "cloud storage and replaces with a URL before sending to the agent. Never base64-encode.",
57
54
  "",
58
55
  "REQUIRED FIELDS:",
59
- "Always check the agent's input schema (via get_agent) before calling run_agent.",
60
- "Include ALL required fields in the first attempt to avoid validation errors.",
56
+ "Always inspect the agent's input schema via get_agent() before calling run_agent(). Include ALL required",
57
+ "fields in the first attempt — validation errors cost a round trip.",
58
+ "",
59
+ "WALLET HYGIENE:",
60
+ "- wallet_status shows per-chain USDC balance and the active network (mainnet vs testnet).",
61
+ "- To create or import a crypto wallet: wallet_setup({ action: \"create\" }) or { action: \"import\", key }.",
62
+ "- NEVER delete or rotate keys programmatically. Direct users to edit ~/.agentwonderland/config.json or ~/.ows/ manually.",
61
63
  ].join("\n"),
62
64
  });
63
65
  // Register tools
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@agentwonderland/mcp",
3
- "version": "0.1.27",
3
+ "version": "0.1.28",
4
4
  "type": "module",
5
5
  "description": "MCP server for the Agent Wonderland AI agent marketplace",
6
6
  "bin": {
package/src/index.ts CHANGED
@@ -31,40 +31,42 @@ export async function startMcpServer(): Promise<void> {
31
31
  },
32
32
  {
33
33
  instructions: [
34
- "Agent Wonderland — AI agent marketplace. Discover, run, and rate agents.",
34
+ "Agent Wonderland — on-demand AI agent marketplace with autonomous per-call payment. Discover, run, rate.",
35
35
  "",
36
- "FILE HANDLING:",
37
- "Pass local file paths directly as input values (e.g. image: \"/path/to/photo.jpg\").",
38
- "The MCP auto-uploads to cloud storage and replaces with a URL. Never base64-encode.",
36
+ "PURPOSE:",
37
+ "You are an AI agent calling this MCP to extend your capabilities. Instead of integrating N APIs with N keys,",
38
+ "you pay specialist agents per call in USDC. This works headlessly in automations (n8n, cron) and interactively",
39
+ "(Claude Code, Desktop). Favor one-shot `solve()` for autonomous use; prefer `search_agents()` + `run_agent()`",
40
+ "when the caller wants to pick a specific agent.",
39
41
  "",
40
42
  "WORKFLOW:",
41
- "1. search_agents() or solve() — find agents for the task",
42
- "2. get_agent() — check required input fields before running",
43
- "3. run_agent() or solve() with ALL required fields",
44
- "3a. If the agent offers discounted credit packs, use buy_agent_credit_pack() and list_agent_credit_packs() when helpful",
45
- "4. If no payment method is set up, run_agent returns a setup page to connect a credit card — present it to the user",
46
- "5. Ask user to rate or tip after a successful run",
43
+ "1. solve(intent, input, budget) — one call: find best agent + pay + run. Use when the task is clear.",
44
+ " OR: search_agents() → get_agent() to inspect schema → run_agent() with required fields.",
45
+ "2. If the agent returns status 'processing', poll with get_job(). Async runs resolve automatically.",
46
+ "3. After a successful run, rate_agent() and optionally tip_agent() if the result was useful.",
47
+ "4. Use list_jobs() to recover state across sessions (it checks every configured wallet).",
47
48
  "",
48
- "PAYMENT:",
49
- "- Crypto wallets (Tempo USDC, Base USDC, Solana USDC) are the core supported rails.",
50
- "- Saved cards can also be used, but card-backed MPP depends on Stripe Shared Payment Token availability in the current environment.",
51
- "- Card and crypto are SEPARATE payment methods. Card charges a credit card; crypto sends USDC on-chain.",
52
- "- When a card is configured, it becomes the default payment method. Use wallet_status to confirm whether `Card MPP: ready` before relying on it.",
53
- "- run_agent() and solve() auto-detect the default compatible payment method when pay_with is omitted.",
54
- "- Include pay_with explicitly when you need a specific rail or want deterministic control over the payment method used.",
55
- "- Use wallet_status to see the exact payment methods the user has configured before calling a paid tool.",
56
- "- Payment is automatic once configured. Users are never charged for failed runs.",
57
- "- Do NOT ask the user to set up payment before they try to run an agent. Let them explore freely.",
58
- "- If a specific payment method fails, report the error clearly. Do NOT silently fall back to a different method.",
59
- "- When payment fails, suggest alternatives using wallet_status.",
49
+ "PAYMENT (crypto-only right now):",
50
+ "- Supported rails: Tempo USDC, Base USDC, Solana USDC. Card is temporarily disabled pending Stripe SPT approval.",
51
+ "- Tempo and Base share one EVM wallet key. Solana uses a separate ed25519 key. One OWS wallet can manage both.",
52
+ "- If pay_with is omitted, the MCP auto-selects a compatible configured rail. Pass pay_with explicitly",
53
+ " (tempo | base | solana | wallet-id) for deterministic behavior.",
54
+ "- Payment is automatic: on a 402 challenge the MCP signs on-chain, submits, then retries. Failed runs are refunded.",
55
+ "- If a specific rail fails, surface the real reason — do NOT silently retry with a different method.",
56
+ "- Headless/automation: set wallet_set_policy() to cap max_per_tx and max_per_day so a runaway loop can't drain funds.",
60
57
  "",
61
- "MANAGING PAYMENT METHODS:",
62
- "- To remove a card: wallet_setup({ action: \"remove-card\" })",
63
- "- To remove or modify a crypto wallet: NEVER do this programmatically. Direct users to edit their config files manually (~/.agentwonderland/config.json or ~/.ows/). We do not handle private key deletion or modification to avoid any risk of key loss.",
58
+ "FILE HANDLING:",
59
+ "Pass local file paths directly as input values (e.g. image: \"/path/to/photo.jpg\"). The MCP auto-uploads to",
60
+ "cloud storage and replaces with a URL before sending to the agent. Never base64-encode.",
64
61
  "",
65
62
  "REQUIRED FIELDS:",
66
- "Always check the agent's input schema (via get_agent) before calling run_agent.",
67
- "Include ALL required fields in the first attempt to avoid validation errors.",
63
+ "Always inspect the agent's input schema via get_agent() before calling run_agent(). Include ALL required",
64
+ "fields in the first attempt — validation errors cost a round trip.",
65
+ "",
66
+ "WALLET HYGIENE:",
67
+ "- wallet_status shows per-chain USDC balance and the active network (mainnet vs testnet).",
68
+ "- To create or import a crypto wallet: wallet_setup({ action: \"create\" }) or { action: \"import\", key }.",
69
+ "- NEVER delete or rotate keys programmatically. Direct users to edit ~/.agentwonderland/config.json or ~/.ows/ manually.",
68
70
  ].join("\n"),
69
71
  },
70
72
  );