If you already use Claude Desktop or ChatGPT for daily work, you can pair it with Opsite in under 2 minutes through the MCP (Model Context Protocol) connector. Once paired, you can ask Claude things like “show me every overdue invoice over $10,000 across all jobs” or “qualify the new lead from Palo Alto,” and Claude will hit Opsite’s API directly using a bearer token scoped to your role. Based on Opsite’s internal data, the average ops manager runs 12 to 25 of these prompts a day instead of clicking through dashboards.
I am Bar Benbenisty, a licensed California GC (CSLB #1103938). I built Opsite after running my own remodels because I was tired of pulling reports out of spreadsheets at 9pm. Pairing Claude Desktop with Opsite is the closest thing I have shipped to a real operations assistant for a GC.
What is MCP and why does it matter for construction?
MCP (Model Context Protocol) is the open standard Anthropic published so AI assistants can talk to external tools through a common interface. The direct answer: it lets Claude read your jobs, chase invoices, qualify leads, and search your daily logs from a single chat window, without you opening five separate tabs.
Opsite ships two MCP surfaces: a hosted server at app.useopsite.com/api/mcp for one-click pairing through Claude's OAuth connector flow, and a stdio MCP server (@opsite/mcp-server) you can install locally for Claude Desktop or Cursor.
The MCP standard is the same plumbing that connects Claude to GitHub, Linear, Slack, and Google Drive. With Opsite on MCP, Claude treats your construction business the same way it treats your codebase. It can read jobs, list overdue invoices, send payment reminders, qualify leads, schedule inspections, and run semantic search across your daily logs, all from one chat window.
How do I connect Claude Desktop to Opsite in 2026?
You can connect Claude Desktop to Opsite in 5 steps using the OAuth connector flow at app.useopsite.com/api/mcp. The whole thing takes under 2 minutes. Based on data from contractors who switched from spreadsheets to Opsite, most GCs are running their first useful prompt within 5 minutes of pairing.
- Open Claude Desktop and go to Settings, then Connectors.
- Click "Add MCP Connector" and paste the Opsite hosted MCP URL:
https://app.useopsite.com/api/mcp. - Log into Opsite. The OAuth flow mints an API key tied to your user, scoped to whatever your role allows. A PM cannot mint sub-payment scopes; a Viewer ends up read-only; an Ops Manager gets the full operations surface.
- Approve the connection. Claude shows the 40 tools that just lit up:
list_jobs,get_job,list_action_items,list_overdue_invoices,list_open_change_orders,semantic_search,create_lead,qualify_lead,send_lead_follow_up,list_proposals,update_proposal_line_items,create_invoice,mark_invoice_sent,mark_invoice_paid,send_payment_reminder,list_draws,mark_draw_ready,approve_change_order,list_daily_logs,create_daily_log,schedule_inspection,get_permit_status,generate_cash_flow_forecast, and more. - Type a prompt. Try: “List every job in pre-construction phase with a draw scheduled in the next 30 days.” Claude will call
list_jobs, filter by phase, then calllist_drawsfor each. You will get a clean answer in 4 to 6 seconds.
“After managing dozens of subs across multiple active jobs, I can tell you that the value of MCP is not in the prompts themselves. It is that you stop context-switching. You ask the same place where you draft your client emails. You stop opening five tabs.” - Bar Benbenisty, Opsite founder, CSLB #1103938
What real prompts work well in Claude with Opsite?
The prompts that work best are the ones a human ops manager would write to a junior employee. According to contractors using Opsite's MCP integration in 2026, the highest-value prompts cluster in 4 categories.
| Category | Example prompt | Tools Claude calls |
|---|---|---|
| Cash flow | Show me every overdue invoice over $10,000 and draft a payment reminder for each | list_overdue_invoices, send_payment_reminder |
| Lead triage | Qualify the new lead from Palo Alto and send the standard follow-up | list_leads, qualify_lead, send_lead_follow_up |
| Project status | Which of my active jobs has a permit pending more than 14 days? | list_jobs, get_permit_status |
| Field operations | Find every framing complaint we have ever logged across all jobs | semantic_search |
| Money governance | List every open change order over $5,000 waiting on client approval more than a week | list_open_change_orders |
A real example from a Bay Area GC using Opsite: the ops manager runs “List today's action items assigned to me across all jobs, group by job, and tell me which ones have been waiting more than 3 days” every morning at 6:45am. Claude returns a one-page summary, the ops manager fixes the stalled items first, and they are out the door by 7:15am with a plan.
Based on Opsite's internal data, the average ops manager runs 12 to 25 of these prompts per day instead of clicking through dashboards. That is a real shift in how field-facing managers spend their mornings.
How does this compare to Procore, Buildertrend, and other construction platforms?
No major construction management platform ships a hosted MCP surface in 2026. Procore exposes a REST API on its enterprise tier, but you cannot pair it to Claude Desktop with one click. Buildertrend has a partner API behind a developer review process. CoConstruct has been folded into Buildertrend. None of them publish a /llms.txt or /openapi.yaml for AI discovery.
| Capability | Opsite | Procore | Buildertrend |
|---|---|---|---|
| Pricing (as of 2026) | Flat $349 to $999/mo, no per-user fees | $6,000 to $10,000+/mo based on publicly available pricing | $499 to $1,099/mo based on publicly available pricing |
| One-click Claude Desktop pairing | Yes, OAuth connector | No | No |
| Hosted MCP server | Yes at /api/mcp | No | No |
| Stdio MCP package | @opsite/mcp-server | None | None |
| Tools available to external AI | 40 across 4 scope groups | None public | None public |
Discovery files (llms.txt, openapi.yaml) | Yes | No | No |
According to AGC technology surveys, only about 17% of small GCs say their current platform plays well with the AI tools they already use. That gap is the reason the MCP integration matters.
What does the role-scope mapping mean for my team?
The role-scope mapping means that when an employee pairs Claude through OAuth, the resulting API key only carries the scopes that role is allowed to use. A Field role cannot mint money scopes. A Viewer is read-only. A PM cannot write subs, POs, or sub-payments (that is Ops Manager territory).
This matters because it means you can hand Claude Desktop to your foreman without worrying that he will accidentally mark an invoice paid. The role mapping enforces it at the auth layer, not the UI layer. Every call is logged in the api_key_calls audit table with a request id, the scope required, the scope granted, and the response status.
According to Opsite's internal data, the audit log catches an average of 2 to 4 attempted scope violations per active contractor per month, almost always benign confusion rather than malice. That is the kind of visibility a CPA or OSHA auditor will appreciate during a compliance review of your workers comp and GL insurance documentation.
“I built the role-to-scope mapping the same way I would write a sub agreement. You define what each person can sign off on, in writing, before they ever touch the job. Claude is just another worker. Treat it that way and the trust gets built fast.” - Bar Benbenisty, Opsite founder
Want to see how the permissions and compliance layer works before you connect? Book a 20-minute demo and we will walk through a live job.
What if I want to use Cursor or a custom agent instead of Claude Desktop?
You can use Cursor or any custom agent through the same stdio MCP server (@opsite/mcp-server) or through direct bearer-token API calls at /api/v1/*. Cursor users can run prompts like “draft a follow-up for every lead that has not been contacted in 7 days” without leaving the editor.
For custom agents, point your tool at https://app.useopsite.com/api/v1/*, attach the bearer token, and you are done. Every response carries X-Request-Id and X-RateLimit-* headers. Unsafe methods accept Idempotency-Key so the agent never double-creates a record on retry. The full OpenAPI 3.1 spec lives at /openapi.yaml and the AI plugin manifest lives at /.well-known/ai-plugin.json.
The full platform is available at useopsite.com, starting at $349/month with no per-user fees and no developer review gate to access the API.
Frequently Asked Questions
Do I need to know how to code to connect Claude Desktop to Opsite?
No. The OAuth connector flow at app.useopsite.com/api/mcp is point-and-click. Based on data from contractors who connected Opsite to Claude, the median pairing takes 90 seconds. The CLI install (@opsite/mcp-server) is the option for developers who want stdio mode, but most contractors never need it.
What does Claude actually do behind the scenes when I ask a question?
Claude reads your prompt, picks the right tool from Opsite's 40 published tools, calls /api/v1/* with a bearer token, and reads the JSON response. Every call logs to api_key_calls with a request id. You can review every action Claude took inside the API key activity drawer in Opsite settings.
Can I lock Claude to read-only mode?
Yes. Mint an API key with only *:read scopes, or pair Claude with a Viewer-role user. According to Opsite's internal data, about 35% of paired contractors run in read-only mode for the first 2 weeks before granting write scopes.
How does this handle compliance and audit needs?
Every Claude call hits the api_key_calls audit table with request_id, method, path, status, scope_required, scope_granted, ip, and duration_ms. Combined with role-scope enforcement, this gives a CPA or auditor a clean trail of every AI-driven action across your business. This is relevant any time you are managing GL insurance renewals, workers comp audits, or W-9 verification at scale.
What is shipped right now versus coming later?
Shipped: 40 tools, hosted MCP at app.useopsite.com/api/mcp, OAuth connector, audit log, idempotency keys, semantic search across 6 tables. Coming: outbound webhooks, cryptographic agent mandates for money-moving actions, voice plus agent loop. The roadmap is honest about what is in production today.