MCP Server
BETAThe Findable MCP server lets MCP-compatible agents answer questions and perform analysis against the buildings, documents, tasks, SharePoint sync state, and red flag data you already have access to in Findable.
Server URL
Use the production URL when connecting an external agent:
https://mcp.findable.aiDevelopment and staging environments use https://mcp.findable.dev.
Connect an Agent
- Open the integrations, connectors, or MCP settings in your AI agent.
- Add a custom connector and paste the Findable MCP server URL.
- Sign in to Findable when prompted.
- Authorize the agent to access your Findable account.
Run this command from your terminal:
claude mcp add --transport http findable https://mcp.findable.aiMicrosoft Copilot Studio
Copilot Studio connects through its native MCP connector, which handles the OAuth sign-in for you.
- In Copilot Studio, open your agent and go to Tools > Add a tool > Model Context Protocol.
- Choose the option that discovers the server from its URL (labelled Dynamic discovery in current builds) and paste
https://mcp.findable.ai. - Create the connection. A Findable sign-in window opens — sign in and choose Allow. Let the window finish redirecting back to Copilot Studio on its own; closing it early leaves the connection incomplete.
- Add the connected tools to your agent and publish.
The Findable MCP server is a public OAuth client: it requires PKCE and never issues a client secret. Power Platform’s “Generic OAuth 2” custom connector requires you to enter a client ID and secret, so that path cannot be completed. Use the native MCP connector above instead.
Each user connects with their own Findable account and sees only what that account can access. Single sign-on is not required. Copilot Studio limits how much a tool may return, so Findable’s search and folder tools reply with compact result lists — the agent then calls get_document to attach the one file you picked.
Access and Security
The MCP server uses OAuth 2.1 with PKCE. Access is tied to your Findable user account, and every tool call is checked against the same permissions as the Findable app.
- Agents only see buildings and documents your user can access.
- Access is in public beta. Your organisation must grant MCP access directly or through a group, or you must have customer administration access.
- You can revoke connected agents from Findable settings.
- 26 tools are read-only. The only write tool starts red flag regeneration.
In the Findable app, go to Settings > AI agents to copy the environment-aware server URL, view connected agents, or revoke access.
Available Tools
The server exposes 27 Findable-specific tools for resource discovery, document search, Q&A, task lookup, SharePoint status, and red flag reporting.
Account and navigation
whoamifind_resourcesget_customerget_buildingget_portfolioget_projectget_collectionLists and discovery
list_buildingslist_portfolioslist_projectslist_collectionslist_userslist_integrationsTasks
list_tasksget_taskDocuments and Q&A
search_documentslist_folder_childrenget_documentget_document_metadataask_findableSharePoint
list_sharepoint_documentsget_sharepoint_sync_statuslist_sharepoint_subscriptionsRed flag
get_red_flaglist_red_flag_itemsget_red_flag_itemgenerate_red_flagWRITECompatible hosts
Findable works with standards-compliant remote MCP clients, including Claude web, Desktop and Code, ChatGPT connectors, and Microsoft Copilot Studio's native MCP connector. Interactive result widgets appear only in hosts that support MCP Apps; all other hosts receive the same data as text.
Interactive Results
In hosts that support MCP Apps — such as Claude — many tools return an interactive widget alongside their text answer instead of a plain list. In hosts without widget support, the same tools return clean text and the agent behaves exactly as before.
- Document search renders as a grid; selecting a result opens the actual document, not just a thumbnail.
- Buildings and portfolios render as cards you can drill into for detail, then search their documents or ask a scoped question.
- Red flag reports render as compliance cards you can drill from theme to finding to the evidence documents behind each finding.
- Widgets follow your Findable language preference and match the host's light or dark theme.
Every action inside a widget — opening a document, drilling into a building — is a fresh tool call, re-checked against your Findable access just like the agent's own calls. Widgets never surface data you could not already reach.
Recommended Flow
Agents should resolve names before fetching detail. Start with whoami to inspect access, use find_resources to resolve customers, buildings, portfolios, projects, or collections, then call the specific getter, list, search, or Q&A tool needed for the task.
MCP clients cache tool schemas at session start. If a newly released Findable capability is missing, restart the agent connection so it can reload the current tool list.
Related Docs
- API authentication - API keys and session tokens for direct API integrations
- Ask endpoint - Build your own Q&A flow without MCP
- Search endpoint - Query Findable documents directly