Tools
02 / 03

Five tools, one of them writes

Four tools read public data. The fifth is the full agent: it prepares buys, sells, launches and fee claims as unsigned transactions, and with a read-write key it can also create scheduled orders.

Updated

The read tools

  • vectr_get_price
    Arguments
    —
    Returns
    One line of text: the native coin's USD price
  • vectr_get_tokens
    Arguments
    limit? (1–200, default 10)
    Returns
    Top tokens by market cap, trimmed to the fields an agent needs
  • vectr_search_token
    Arguments
    query
    Returns
    Tokens matching a name or symbol, same trimmed fields
  • vectr_get_token
    Arguments
    address (0x + 40 hex)
    Returns
    One token's detail plus curve progress (%) and its vectr.bot page

None of these need VECTR_API_KEY. They wrap the public REST endpoints: the list tools trim each token to the fields an agent needs, and vectr_get_token adds curve progress and the token's page URL.

vectr_agent

  • prompt
    Type
    string (required)
    Notes
    Natural-language instruction — research, trade prep, launches, orders, portfolio reads, fee claims
  • model
    Type
    string?
    Notes
    Model override. The backend default is claude-sonnet-5, and an unknown id falls back to it

Under the hood the tool creates an /agent/job, polls it every 1.5s, and returns the finished response, or gives up after 180 seconds and tells you the job id to poll later. When the turn produced a transaction, the answer ends with the unsigned transaction and a note that it has not been signed. The tool cannot broadcast.

The boundary is the point

An MCP client can ask the agent to prepare anything it can describe — but signing stays in the wallet owner's hands. If a workflow needs fully automated execution, that's the read-write API key path in API → Authentication, not MCP.