MCP
Ask your assistant about your AI visibility data.
A read-only MCP connector for one project at a time. Your assistant can read the saved data its tools expose, and nothing more.
What it is
A read-only connector over Streamable HTTP.
The connector lets an MCP client read one project’s saved prompts, answers, citations, daily metrics, weekly digests, brand reasons, objections, attributes and fact checks, and run Explorer queries and saved views.
It uses the same project keys, permissions and plan access as the customer API. Your client connects over Streamable HTTP and sends the key in an Authorization header.
Connect
Four steps, one key per project.
- 1
Open the client project in Settings → API keys. An owner or editor can create a key; a viewer cannot.
- 2
Select only the read permissions your assistant needs. The table below names the scope each tool requires.
- 3
Choose an expiry of 30, 90 or 365 days. The secret is shown once, so copy it straight into your client’s secure credential settings, not into a chat message.
- 4
Add a Streamable HTTP server with the MCP connection URL shown in settings, keeping the trailing slash, and set the header
Authorization: Bearer YOUR_KEY.https://api.discoveredby.ai/customer-api/v1/projects/PROJECT_ID/mcp/
There is no OAuth sign-in step. A client that only offers browser sign-in and cannot send a custom header cannot use this connector. Revoking the key stops future requests, and revocation is permanent.
Tools
The 14 tools and the scope each needs.
The report, brand reasons, objections, attributes and fact check tools also need that feature on the project owner’s plan.
| Tool | Scope | What it reads |
|---|---|---|
| list_prompts | prompts:read | Current prompts, including paused prompts and their current labels. |
| list_answers | answers:read | Completed answers with collection provenance, the target location and how it was sent to the engine, the persona and the language variant; previews stop at 8,000 characters. |
| read_answer_text | answers:read | The full saved text of one answer, read in bounded, version-checked chunks. |
| list_citations | citations:read | Saved citation records with at most 2,000 snippet characters, not fetched page contents. |
| read_daily_metrics | metrics:read | Daily counts, brand visibility and own-domain citation rate by recorded collection channel. |
| list_reports | reports:read | Saved weekly digests for completed Monday to Sunday UTC weeks; also needs weekly-report access. |
| read_report | reports:read | One exact saved digest from list_reports, with its four saved readings; also needs weekly-report access. |
| list_brand_reasons | reasons:read | Strength and weakness reasons a completed answer gave for a tracked brand, each with its category and a quoted excerpt. |
| list_objections | objections:read | Objections engines raised when asked why a buyer might not choose the brand or a tracked competitor, with rank, group, quote and cited sources. |
| list_attributes | attributes:read | What engines associate with each brand, or which brands they name as best known for an asked attribute, from succeeded attribute studies. |
| list_fact_checks | facts:read | Where AI answers supported or contradicted the brand’s approved facts, with the verdict, review, quote and cited sources. |
| query_metrics | explorer:read | Computes Explorer metrics for the project now, the same numbers the app’s Explorer shows. |
| list_explorer_views | explorer:read | The project’s saved Explorer views with their stored queries and charts, and whether each needs repair. |
| run_explorer_view | explorer:read | Runs a saved view’s query now and returns the view and its result, as query_metrics would. |
Limits
What to know before you connect.
- Read only. The tools cannot change data, collect new answers or send messages, and every tool is marked read-only to your client.
- One project per connection. Tool arguments cannot select another project, so each client project needs its own key and connection.
- Each tool checks for its own scope when it is called, even though discovery lists every tool.
- Null means unavailable, not zero. Rates with fewer than 30 observations are marked provisional.
- Live reads, not snapshots. Rows can change between calls, and your assistant must follow pagination to get complete results.
- List tools return pages: up to 100 prompts, or up to 25 answers, citations, reports, reasons, objections, attributes or fact checks per call.
- Each key admits 60 authenticated requests a minute, shared with the API. Discovery, initialization and refused tool calls count too.
- query_metrics and run_explorer_view each spend one query from the project’s Explorer budget of 120 a minute, which the app shares.
- The server tells your assistant to treat saved prompt, answer, citation and report text as evidence, never as instructions.
- No OAuth sign-in, no HTTP+SSE transport and no local stdio command. Requests are Streamable HTTP POSTs with bodies up to 64 KiB.
Plans
Which plans include it.
Access follows the project owner’s plan, whoever created the key. The MCP connector and the API are included together.
| Free | Starter | Growth | Pro |
|---|---|---|---|
Recipes
Prompts to paste into your assistant.
Each recipe uses only the tools listed with it, so your key needs only their scopes.
Weekly visibility briefing
Compare the last seven days of brand visibility and own-domain citation rate with the seven before, one collection channel at a time.
read_daily_metrics
Competitor share-of-voice check
See how your brand and each tracked competitor compare on share of voice, visibility and position, against the previous period.
query_metrics
Citation sources report
List the domains AI answers cited over the last 28 days, with your own domain marked and each collection channel kept apart.
list_citations
Why answers favour or caution against each brand
Summarise the strength and weakness reasons AI answers gave for your brand and each tracked competitor, with quoted evidence.
list_brand_reasons
Client report draft
Turn the latest saved weekly digest into a short, client-ready summary you can edit before it goes anywhere.
list_reports, read_report
Run your saved Explorer views
Run each saved Explorer view now and get one line on what each shows. Each run spends one query from the Explorer budget.
list_explorer_views, run_explorer_view
Prompt performance review
Find the prompts where your brand gained or lost the most visibility over the last 28 days.
list_prompts, query_metrics
Answer features scan
See how often each engine’s answers used web search or held shopping results, local businesses, ads, video, images or tables.
query_metrics
Every field and error, documented.
The MCP guide covers pagination, measurement rules and troubleshooting for each status code.