Tool reference
All 24 tools available over MCP and the API
Reading your data
These 19 tools answer questions and change nothing. Every one of them is scoped to your workspace — a key cannot read another customer's project, and asking for one returns the same answer as asking for a project that does not exist.
| Tool | What it returns |
|---|---|
listProjects | List the projects in this workspace with their tracked brand, market and last measurement date |
getProject | Full settings and context for one project: brand, market, tracked engines, categories, and the description read from the client's own site |
getBrandMetrics | One metric for the tracked brand plus the competitors ahead of and behind it |
getEngineBreakdown | The same metric split by which assistant answered |
getBrandMetricsByPrompt | Per-prompt breakdown: which tracked questions the brand shows up in and which it never does |
listAnswers | The full answers the AI assistants gave to tracked prompts, with the sources each one cited |
listEntities | People, organizations, media, places and products named in the assistants' answers, with their aliases, attributes (such as party or office) and how many answers name each one |
getEntityMetrics | How often entities are named, grouped the way you ask - by engine, topic, entity or an attribute such as party |
getVerdictMatrix | For questions that ask for a judgment (who is favored, which is best), what each assistant concluded: a prompt x engine grid of favorite / tossup / no_call / refused, the favored entity and its group (e.g |
getEngineStats | Per-assistant answer statistics: how many answers, how often the assistant showed no answer or refused, average answer length, entities per answer, distinct entities and verdict counts |
getSourceBreakdown | How citations split across source categories and content types |
listCitedSources | Domains or individual pages the models cited when answering, with how often each was used and how often the tracked brand appeared in those same answers |
getSourcePromptImpact | Which tracked prompts pull from one source, and through which engines |
listPublishingOptions | Cited domains where the client can actually buy a publication through our partner, with the partner's cheapest listed price |
getSiteAudit | The latest GEO audit of the client's own website: twenty-two checks in four groups, each with a verdict, what it measured and what to do about it |
listPrompts | The questions being tracked for a project, with their intent and topic tags |
listTopics | Topics for a project - the subjects buyers research |
listBrands | Brands discovered in the answers, with how they are classified and which name variants were merged into them |
getCreditUsage | Credit balance and what has been consuming it, split by tracking versus agent work |
Doing work
These 5 change something, and each returns a plan before it acts.
| Tool | What it returns |
|---|---|
addPrompt | Start tracking a new question for a project |
removePrompt | Stop measuring a tracked question |
updateEntity | Correct entities found in the answers: set or clear an attribute (such as party) on one or many entities, change the type, rename one, or add an alias |
updatePrompt | Set the category, framing (neutral, partisan, branded) or expectsVerdict of one or many tracked prompts |
startAgentTask | Put the Scorra agent to work on something: an article, an outreach email, an analysis, a content plan |
Defaults
When a tool is called without a period, it answers for the last 30 days across all engines. Say the window explicitly when it matters — last 7 days, on Gemini — because a summary that silently changed its window looks like the brand moved.
Discovering this list from code
GET https://app.getscorra.com/api/v1/tools Authorization: Bearer <key>
Returns the same catalogue filtered to your key's scopes, with JSON Schema for every input. That endpoint, this page and the MCP server all read one registry, so they cannot disagree.