Agent Guild MCP tool for ADK¶
Agent Guild provides endpoint preflight, signed reputation evidence, and pricing discovery through a hosted MCP server. Connect it with ADK's MCP toolset to inspect an external agent before deciding whether to delegate work to it.
Use cases¶
- Inspect a delegation target: Compare an endpoint's advertised capabilities with the observations and unknowns returned by a live preflight.
- Check costs before use: Retrieve current paid-operation prices, callable entrypoints, and free alternatives before choosing a trust operation.
- Review evidence limits: Keep unavailable checks and uncertainty visible when explaining whether an endpoint is suitable for a task.
Prerequisites¶
- Set up ADK and model access using the Python or TypeScript quickstart.
-
Install MCP support for the language you use:
The two tools in this example are free and require no Agent Guild account, API key, or wallet. Model-provider usage follows your model provider's pricing. Preflight sends the selected public endpoint URL to Agent Guild's hosted service, which probes that endpoint. Supply public URLs without embedded credentials.
Use with agent¶
The tool filter exposes only endpoint preflight and pricing discovery. It does not expose registration, payment, or paid trust-read tools to the agent.
from google.adk.agents import Agent
from google.adk.tools.mcp_tool import McpToolset
from google.adk.tools.mcp_tool.mcp_session_manager import StreamableHTTPConnectionParams
root_agent = Agent(
model="gemini-flash-latest",
name="agent_guild_preflight",
instruction=(
"Inspect the public agent endpoint the user supplies before delegation. "
"Use guild_preflight and report its observations and unknowns. "
"Use guild_paid_operations when the user asks about trust-operation prices. "
"A preflight result is evidence, not a guarantee or permission to execute."
),
tools=[
McpToolset(
connection_params=StreamableHTTPConnectionParams(
url="https://agent-guild-5d5r.onrender.com/mcp",
timeout=30,
),
tool_filter=["guild_preflight", "guild_paid_operations"],
)
],
)
import { LlmAgent, MCPToolset } from "@google/adk";
const rootAgent = new LlmAgent({
model: "gemini-flash-latest",
name: "agent_guild_preflight",
instruction:
"Inspect the public agent endpoint the user supplies before delegation. " +
"Use guild_preflight and report its observations and unknowns. " +
"Use guild_paid_operations when the user asks about trust-operation prices. " +
"A preflight result is evidence, not a guarantee or permission to execute.",
tools: [
new MCPToolset(
{
type: "StreamableHTTPConnectionParams",
url: "https://agent-guild-5d5r.onrender.com/mcp",
},
["guild_preflight", "guild_paid_operations"],
),
],
});
export { rootAgent };
Run the agent using the quickstart for your language. For example, ask it to inspect a public agent endpoint you operate and explain which checks were inconclusive. Then ask which paid trust operations are available and what each currently costs.
Available tools¶
| Tool | Description |
|---|---|
guild_preflight |
Probe a public agent endpoint and return observed capabilities, evidence, and unknowns. Free without a key. |
guild_paid_operations |
List current paid-operation prices, entrypoints, settlement resources, and free alternatives. This discovery call is free. |
The server also offers signed Agent Passports and paid trust reads. These are
outside this example's tool filter. A signature establishes origin and
integrity, not the safety of an agent's messages or the quality of future work.
Paid reads such as guild_check require the current payment terms or funded
sandbox credits; do not treat them as free preflight calls.