Skip to main content
BestProp.ai

MCP & connectors

Connect AI applications to separate BestProp.ai property research and GST MCP services, with setup guides for ChatGPT, Claude, and Codex.

Connection status

Public documentation needs no sign-in. Use the endpoints shown here when the service is available. ChatGPT app publication and verification in each AI host are separate launch steps.

REST API
Public launch pending
Research MCP endpoint
Public launch pending
GST MCP endpoint
Public launch pending
ChatGPT app
App publication pending
Claude
Host verification pending
Codex
Host verification pending

Use BestProp.ai's MCP connector to search public property reports, read news and blogs, retrieve recent publications and compare projects from an assistant. The same public data is available through the REST API.

No sign in for public reads

The four public tools require no BestProp.ai account, cookie, API key or OAuth sign in. Choose No sign in or the equivalent anonymous option when your host asks how to authenticate. Do not enter database credentials or send website session cookies.

Your assistant account and workspace still control whether external tools are available and whether you can add a custom connection. Enabling a connector does not expose private BestProp records or give it permission to edit content.

MCP endpoints and connections

Use the configured MCP endpoint shown on MCP & connectors. An example is https://api.bestprop.ai/mcp; this example hostname is not evidence of a live service. Check the page's availability and platform status before connecting.

Research and GST require separate connections:

Package Endpoint Discovery
bestprop-ai (BestProp.ai) /mcp Four research data tools, app-only open_research_workspace, one research resource.
bestprop-gst (BestProp.ai GST) /mcp/gst Eleven GST calculation/review tools, app-only open_gst_workspace, one GST resource.

Both connections are anonymous. The GST endpoint does not expose property research and the research endpoint does not expose GST. Separate package identities and endpoints do not claim either listing is currently published.

For a generic MCP client:

  1. Create a Streamable HTTP connection to the exact configured /mcp or /mcp/gst URL for the intended surface.
  2. Run the MCP initialization handshake and complete the initialized notification.
  3. Discover available tools with tools/list and read their input schemas.
  4. Call tools with tools/call, validate structured results and retain disclosures.

Use a maintained MCP client library to manage protocol negotiation and transport headers. Opening /mcp in a browser or sending a normal JSON GET does not perform this handshake. See the official MCP client guide.

Available tools

Tool What it does Input guidance
search_content Find published projects, news, blogs and articles. Use kind=project for city, area, builder or zone filters.
get_content Read a report's public facts, sources and sections. Use an exact slug returned by search; follow nextSectionOffset for remaining sections.
list_recent_content Retrieve newest currently published content. Choose an optional kind and follow the returned continuation cursor.
compare_projects Compare two or three published projects. Supply distinct project slugs in the order you want shown.

Slugs are identifiers, not guessed titles or external URLs. News and blog slugs can contain multiple path segments. Budget, BHK and possession filters are not supported. Tools are read-only; they do not create bookings, contact requests, saved lists or publications.

ChatGPT

Public app listing (currently called a plugin)

When an approved BestProp.ai listing is published, its link appears on this website. Follow that listing to install or connect through ChatGPT and enable it in a conversation. Listing availability, installation and tool permissions are controlled by OpenAI and your workspace.

A deployed MCP endpoint and a public directory listing are separate milestones. OpenAI requires submission, review and publication. A developer connection does not establish public listing availability. OpenAI submission and publication.

Developer preview

OpenAI's current documentation calls the distribution unit a plugin shared by ChatGPT and Codex; this is the branded ChatGPT app integration, not a custom GPT or the retired legacy ChatGPT plugin protocol.

Developer mode is available on the web for Plus, Pro, Business, Enterprise and Education accounts, subject to workspace policy. It is not promised for Free accounts. Official eligibility.

  1. Open Settings → Security and login → Developer mode in ChatGPT.
  2. Open ChatGPT Plugins, select +, and enter BestProp.ai, a description and the qualified remote /mcp URL.
  3. Choose No Authentication for BestProp's public reads; create the connection.
  4. Inspect the four research data tools, app-only workspace launcher and metadata. Start a new conversation and select the connection from the tools/Developer mode menu.
  5. Ask it to search published reports, read an exact returned slug, list recent publications and compare two actual projects. Keep tool arguments, results and UI evidence; do not treat connection creation as complete qualification.
  6. After server metadata changes, open the connection, select Refresh, then start a new conversation and repeat affected cases.

A qualified public HTTPS preview is required for this flow. OpenAI also supports Secure MCP Tunnel for private development, but a tunnel does not replace the public endpoint required for submission. Connection and testing instructions.

Claude

For an individual account with custom connectors available, open Customize → Connectors, choose Add custom connector, enter a name and the configured remote MCP URL, then review authentication and select No sign in. Add the connection and enable it for a conversation.

Custom remote connectors are available on Free, Pro, Max, Team and Enterprise; Free allows one. Team setup requires an Owner or Primary Owner in Organization settings → Connectors → Add → Custom → Web. Enterprise also allows a custom role with Manage access to Libraries. Members then connect individually under Customize → Connectors. Enable the connector using the conversation's + → Connectors menu.

Claude makes remote connections from Anthropic's cloud, even in Claude Desktop and Cowork. A loopback or VPN-only endpoint is not sufficient. A custom connection and a Claude directory listing are separate release paths. See Claude's current custom connector instructions.

Codex

Codex supports remote Streamable HTTP MCP configuration. Its connection and UI behavior require separate qualification; a successful ChatGPT or Claude session does not qualify Codex.

In a supported Codex client, add a Streamable HTTP MCP server with the configured URL. For the CLI, put the following entry in ~/.codex/config.toml (or a trusted project’s .codex/config.toml). Replace the example with the endpoint shown on this website:

[mcp_servers.bestprop]
url = "https://api.bestprop.ai/mcp"

No bearer token or OAuth login is needed for BestProp public reads. Use codex mcp list or the CLI’s /mcp view to inspect the configured server before using it. Check discovered tools in your client. Follow OpenAI's Codex MCP configuration reference for the setup supported by your version and execution environment.

Installing the packaged plugin instead of direct MCP

Direct MCP configuration above is the shortest local Codex setup; it does not install a directory plugin, and ChatGPT web does not read ~/.codex/config.toml. After public approval/publication, search the universal Plugins Directory for BestProp.ai or open the actual listing supplied by the publisher and install it in the supported host. No BestProp public listing is claimed here.

For local package testing, an operator first builds the repository's distribution folder and adds it to a repo/personal marketplace. Current official instructions use the ChatGPT desktop app's Plugins Directory to install/test local plugins; codex plugin marketplace add registers a catalog, not an individual MCP server or a public listing. Repository operators have separate deployment/submission instructions; repository access is not required to use a future published listing.

The authoritative OpenAI package and local marketplace guide describes client-specific availability and the codex plugin marketplace add/list/upgrade/remove commands. Do not assume local marketplace support on every web, cloud or CLI surface.

GST connection and local tools

For GST in ChatGPT or Claude, create a separate BestProp.ai GST connection using https://api.bestprop.ai/mcp/gst on the qualified public origin and choose no authentication. Discover the tools before supplying redacted sample inputs. The GST connection supports calculation, invoice/quotation drafts, qualified calendar reminders, spreadsheet mapping, reconciliation, liability and return review. It cannot read property reports, access the GST portal, file returns or make payments. Matching does not grant ITC entitlement and missing evidence or ledgers must remain visible as qualifications.

For a direct Codex connection, use a distinct server entry:

[mcp_servers.bestprop_gst]
url = "https://api.bestprop.ai/mcp/gst"

When installing complete packages, choose BestProp.ai for research and BestProp.ai GST for GST only after their actual listings are published, or use the corresponding prepared packages in the local bestprop marketplace for qualification. The GST package supplies six skills and scripts/gst-books-cli.mjs, a manually invoked Node 20+ bundle that processes explicitly authorised paths. It never runs automatically. The research package has no GST skills or CLI. A direct MCP entry does not install those skills.

Local GST workspace selection processes selected files in memory. Remote file calls instead pass data through the host platform and BestProp service; only configured trusted HTTPS download origins are allowed, and remote download is disabled by default. Inline small files remain subject to request bounds. Discover the exact input schemas and continuation fields. Do not treat a paginated preview as a complete book or invent file URLs, calculations or output artifacts when a tool is unavailable. The CLI uses original authorised files and confirmed mapping/context; remote tools must also reuse the original file rather than a preview. All outputs are human review drafts or unqualified offline-tool inputs, never government acceptance or filing receipts.

Host qualification must independently exercise the GST workspace, deterministic calculation, calendar coverage warnings, authorised file review and refusal of portal credentials/filing. A prepared test case or local browser fixture is not an executed host journey or reviewer recording.

Current data, evidence and assistant answers

New supported content becomes available after its publication transaction commits. Subsequent REST and MCP requests read the current public records; drafts and unsupported content types stay private. Updates, unpublishing and removal are reflected on subsequent reads without a separate connector import.

An existing assistant answer does not rewrite itself automatically. Ask the assistant to call the tool again when current data is needed. The recent feed is a request-based feed, not a push subscription.

Preserve returned canonical BestProp.ai links, original dates, source attribution, evidence limitations and conceptual-image disclosures. servedAt identifies retrieval time, not the date a price or underlying fact was verified. Unknown facts remain unavailable. Reports do not promise live inventory or a guaranteed price. Hosts without interactive UI support can still use text and structured results.

Property research workspace

Compatible hosts can expose BestProp.ai research as a sidebar or thread panel via the app-only open_research_workspace entrypoint. That entrypoint opens a UI; the four public tools above remain the data capabilities. Installation, enabled tools and interactive UI support are separate.

The workspace starts without a previous tool answer. Search published content or choose recent publications, open a report inside the panel, return to the results and select two or three projects in the order to compare. Report facts, canonical citations, source limits and conceptual-image disclosures remain visible. External report links open through the host, not an embedded website.

Navigation preserves validated public identities, not saved fact snapshots. Reopening or refreshing reads the current publication; removed/private content cannot remain as a previously displayed report or selected context. Only bounded selected public titles, identities and canonical URLs are offered as model context. Discuss requires an explicit click and a supported host action; the workspace never sends a message automatically.

Unsupported hosts retain text/structured tool results. A local bridge exercise does not establish ChatGPT web/desktop, Codex CLI/desktop/cloud or Claude compatibility. Provider account policies and available extension APIs determine whether entrypoints, restored navigation, model context and Discuss are usable. The publisher qualifies each surface before advertising support.

After tool/UI metadata changes, refresh/reconnect through the installed client's supported control, start a new conversation and repeat affected journeys. Remove the plugin/connection through that same client's installed-plugin or connector controls; removing a local marketplace alone is not proof that every client installation or direct MCP entry was removed.

Future account actions

Saved lists, personal history, private research and write actions are outside this public connector. Such capabilities would require a separate authenticated extension with OAuth, scoped authorization and explicit user permissions. Public-read access does not authorize those actions.

Reference

Use the configured service's OpenAPI schemas for REST integrations and MCP tool discovery for assistant integrations. Reading these pages alone does not give an AI application network access or automatically connect its tools.