Agents

Dieser Inhalt ist noch nicht in deiner Sprache verfügbar.

Agents are purpose-driven tool compositions that bundle tools from multiple providers for a specific task. They replace the simpler “Groups” concept from v2.0.0 with full manifest definitions including model binding, system prompts, and testable behavior.

Agent vs Provider

Aspect	Provider	Agent
Scope	Single API source	Multiple providers combined
Prompts	Model-neutral	Model-specific with `testedWith`
Model	None (any model can use)	Bound to specific LLM
Purpose	Expose data	Accomplish tasks
Tests	Per-tool deterministic	Tool selection + content assertions

Manifest Format

Each agent is defined by a manifest.json file in the catalog’s agents/ directory.

{
    "name": "crypto-research",
    "version": "flowmcp/3.0.0",
    "description": "Cross-provider crypto analysis agent",
    "model": "anthropic/claude-sonnet-4-5-20250929",
    "systemPrompt": "You are a crypto research analyst...",
    "tools": [
        "coingecko-com/tool/simplePrice",
        "etherscan/tool/getContractAbi",
        "defillama/tool/getProtocolTvl"
    ],
    "tests": [
        {
            "_description": "Basic token lookup",
            "input": "What is the current price of Ethereum?",
            "expectedTools": ["coingecko-com/tool/simplePrice"],
            "expectedContent": ["price", "USD"]
        }
    ]
}

Required Fields

Field	Type	Description
`name`	string	Agent name (kebab-case)
`description`	string	Human-readable purpose
`version`	string	Must be `"flowmcp/3.0.0"`
`model`	string	Target LLM (OpenRouter format with `/`)
`systemPrompt`	string	Agent persona and behavioral instructions
`tools`	string[]	Tool IDs in `namespace/type/name` format
`tests`	array	Minimum 3 test cases

Optional Fields

Field	Type	Default	Description
`maxRounds`	number	10	Maximum LLM interaction rounds
`maxTokens`	number	4096	Maximum tokens per response
`prompts`	string[]	[]	Relative paths to prompt files
`sharedLists`	string[]	[]	Required shared list names
`inputSchema`	object	—	JSON Schema for agent input

Agent Tests

Each agent must have at least 3 tests. Tests validate tool selection (deterministic) and optionally content assertions.

{
    "_description": "Cross-provider DeFi analysis",
    "input": "Compare TVL of Aave on Ethereum vs Arbitrum",
    "expectedTools": ["defillama/tool/getProtocolTvl"],
    "expectedContent": ["TVL", "Ethereum", "Arbitrum"]
}

expectedTools — Deterministic: the agent must call exactly these tools
expectedContent — Partial: the response should contain these strings

Validation Rules

Code	Rule
AGT001	`manifest.json` must exist and be valid JSON
AGT002	`name` must be kebab-case
AGT003	`version` must be `"flowmcp/3.0.0"`
AGT004	`model` must use OpenRouter format (contains `/`)
AGT005	`tools[]` must not be empty
AGT006	Each tool ID must be valid (`namespace/type/name`)
AGT007	`tests[]` must have minimum 3 entries
AGT008	Each test must have `_description`, `input`, `expectedTools`

CLI Commands

# Import an agent from a catalog
flowmcp import-agent crypto-research

# The agent's tools are activated locally
flowmcp list