Public API + MCP
Ask Cribber from your AI assistant
Cribber compares agents and teams using 12 months of local listing activity, including active listings, homes with recorded sales, and other off-market listings. Each monthly leaderboard uses consistent geographic boundaries and crediting rules, with supporting listings you can inspect. No Cribber account, API key or login is required.
Read the data & AI access guide for ranking definitions, edition dates and state → city → neighborhood navigation.
Read public content with any AI assistant
AI search, retrieval, citation and training crawlers are welcome on Cribber’s public content. See our public content and AI access permission. No account, JavaScript, API key or MCP connection is needed to read a leaderboard.
Open the HTML city and neighborhood directory, or use the compact JSON directory and plain-text directory. The JSON and plain-text versions return exact page links, state filters and pagination. Follow the returned state, city and neighborhood links when a search engine has not surfaced a place; no XML or Markdown parser is needed.
Each leaderboard has a complete Markdown table at its URL plus .md, and structured data at .json. Both use the same published edition, rankings and evidence as the website. Use city and state to distinguish places with the same name.
For a small reading page, append .md?limit=25 or .json?limit=25 and follow the next-page links. Each page states its position, total cohort, dates and metric. Equal ranks can continue on the next page. Omit limit to download the complete edition.
A public MCP address does not automatically connect ordinary ChatGPT or other search assistants. Use the direct page or downloads for web retrieval; connect MCP explicitly for interactive structured queries.
Connect Claude Code
Run this once in your terminal:
claude mcp add --transport http --scope user cribber https://cribber.ai/mcp
Then start a new Claude Code conversation and ask: “Use Cribber to find top agents in Boston’s South End.” Follow up with “Only condos between $1 million and $2 million,” “Compare the top three by asking volume,” or “Show me the listings behind that result.”
Other MCP clients can connect to https://cribber.ai/mcp with Streamable HTTP and no authentication. Your assistant keeps the conversation context; Cribber supplies public evidence.
What you can ask
- Find supported cities and areas, including newly published cities.
- Read the website’s published monthly leaderboards and their historical editions.
- Refine observed listing activity by area, property type, asking price and price per square foot.
- Find agents by name, read public profiles, and inspect known inventory or monthly listing evidence.
Use the API directly
View the live tool catalog and input schemas. Send a JSON POST to /api/v1/agents/query using the same tool names and arguments as MCP:
curl https://cribber.ai/api/v1/agents/query \
-H 'Content-Type: application/json' \
-d '{"name":"find_locations","arguments":{"query":"South End"}}'Use the exact returned market, area and geography IDs for subsequent calls. List public markets. Ambiguous names need a city or state; unpublished locations are unavailable.
Understand the results
Published monthly leaderboards are frozen editions. Filtered searches use observed trailing-12-month listing activity and may differ from those editions. Dated snapshots keep their original dates. A home’s history does not end when it leaves the market. Cribber keeps qualifying past listings in the comparison, helping you see an agent’s broader listing record over time. Homes Listed counts distinct credited homes; Asking Volume adds their selected asking prices. Named teams can appear alongside individuals.
Use the source links, reporting window, coverage and paging fields returned with each result. Active, sold and inactive records describe different stages of the listing record; sold status uses qualifying recorded-sale evidence. Empty results and temporarily unavailable data mean different things.
Access and limits
All seven tools are read-only. Queries read stored evidence and do not run paid research, publish cities or send messages. Anonymous requests are limited to 60 per minute per hosting-observed IP; HTTP 429 includes a retry delay. Existing underlying endpoints may apply additional limits. Broad results are paginated.