Workflows

These are the patterns your agent will use most. Each shows the tool sequence and the arguments that matter; ids from earlier steps feed the later ones.

Name a new product, end to end

The core loop: create a project, add candidates, analyze one, and compare.

1. create_project      { name: "Aurora", industry_category: "Technology" }
                       → project_id, public_id
2. add_names           { project_id, names: ["Aurora", "Lumen", "Vela"] }
                       → created rows with name_id for each
3. run_full_analysis   { project_id, name_id, name: "Aurora" }
                       → run_id
4. get_analysis_status { run_id }   ← poll until status = "completed"
5. get_score_breakdown { name_id }  ← read the score
   get_availability     { name_id } ← read domains/socials
   get_name_meaning     { name_id } ← read meaning/etymology
6. compare_names       { project_id }  ← rank everything at once
7. shortlist_name      { name_id }     ← keep the winners

run_full_analysis returns right away. Poll get_analysis_status with the run_id until it reports completed, then read each result with its own tool.

Quick availability sniff test

Before committing a name to a project, check it ad-hoc — no project, no credits spent:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "check_name_availability",
    "arguments": {
      "name": "Aurora",
      "tlds": [".com", ".ai", ".io"],
      "platforms": ["twitter", "github"]
    }
  }
}

To persist the result against a saved candidate instead, include both project_id and name_id — that runs the full check and spends credits.

Screen a shortlist for risk

Once you have candidates you like, screen them before presenting:

for each shortlisted name_id:
  check_global_risk       { project_id, name_id, name }   ← cultural / offensive risk
  check_trademark         { project_id, name_id, name, classes: [9, 42] }
  check_brand_similarity  { name }                        ← copycat check

classes are NICE trademark classes (1–45); omit them to let PowerNames pick classes from the project's industry.

Rate and select

Blend the machine score with human judgement, then lock in a choice:

rate_name       { name_id, rating: 9 }        ← record your gut score (1–10)
compare_names   { project_id, name_ids: [...] } ← final side-by-side
shortlist_name  { name_id }                    ← or leave as-is

Undo a mistake

Every mutation returns an auditId. If you added the wrong names or shortlisted the wrong candidate, reverse it:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "undo",
    "arguments": { "auditId": "AUDIT_ID_FROM_A_PRIOR_RESULT" }
  }
}

Analyses and anything that spent credits can't be undone — only reversible mutations qualify.

Learn the methodology first

For deeper naming work, have your agent read the methodology guide before it starts, so its scoring and recommendations match how PowerNames thinks:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "read_skill",
    "arguments": { "skill": "naming-methodology" }
  }
}