Read-only research
04 / 04

Research needs no write access

Most of what an agent does is read — prices, portfolios, order books, fee balances. You can run a fully useful agent that is physically incapable of moving funds.

Three levels of read

  • No credentials
    What it can do
    Every public endpoint: tokens, quotes, candles, streams, config
    What it can't
    Agent jobs, anything wallet-scoped
  • Read-only API key
    What it can do
    Agent jobs — portfolio reads, market research, order listing — all planning, no signing
    What it can't
    sign-and-submit returns 403, always
  • Read-write key
    What it can do
    Everything, including broadcasting
    What it can't
    Nothing — that's the point of the other two

The MCP server is keyless-friendly: four of its five tools work with no key at all, so a research agent in Claude or Cursor can query the market without ever touching a credential.

Prompts that stay read-only

"what's my portfolio worth right now"
"which tokens I launched earned the most in fees"
"show my open orders and when they trigger"
"what graduated this week and at what market cap"
"compare 0x… and 0x… — price, volume, holders"

Every one of these is a read under the hood — the agent's market-data and wallet-read tools don't touch the signing path, and a read-only key makes that structural rather than behavioral.

Building one

# profile → API keys → create with readOnly: true
# then use it anywhere:
curl -X POST https://api.vectr.bot/agent/job \
  -H "X-API-Key: vectr_…" -H "Content-Type: application/json" \
  -d '{"prompt":"summarize my positions"}'

Read-only is the right default

A research bot's key can leak without consequences worse than a rate limit. Start read-only, add write scope to a separate key only when something genuinely needs to execute.