LLM Integration · CRM & ERP
Salesforce, HubSpot, Pipedrive, Shopify, SAP, DATEV, Lexware — we connect your existing tool with Claude, ChatGPT or your own bot. You ask *which deals have 80%+ probability this month* and you get the answer. With permissions, sandbox test and audit log.
What you get
We don't just wrap your API. We build the full bridge including permissions, sandbox mode and bidirectional sync. So that *can be added* becomes a reliable production setup.
What it's used for
You ask Claude in your daily *which deals have 80%+ probability this month and last update older than 7 days* — Claude calls HubSpot via MCP, filters, answers with a list plus recommendations who to follow up. Saves 15-20 minutes per day for pipeline hygiene.
You ask *which products under 80 EUR in size L are in stock and have sales rising compared to last month* — the AI calls Shopify API and delivers the answer. Plus recommendations which products to push.
Client calls become more productive when you ask Claude during the call *what are the open invoices for client X this quarter*. DATEV data comes via MCP bridge. Read-only, audit log mandatory, GDPR contract with DPA clause.
Developers query CRM data from the IDE (e.g. for customer success features or reporting dashboards), without having to open the Salesforce UI. Saves context switching.
Inquiries from your website bot are automatically sorted into Pipedrive. Bot knows the Pipedrive pipeline stages, creates deals with correct tags and owner assignment. Saves manual CRM maintenance.
How it works
We walk through what your CRM or ERP or shop can do (API availability, endpoints, rate limits, data model). Plus GDPR mapping: which fields are sensitive, which need explicit permission, which don't leave.
If your tool is MCP-capable (HubSpot, Salesforce, Notion, Linear have standard MCPs), we use it. If not (DATEV, Lexware, many ERPs) we build a custom MCP. Plus permission layer on MCP tool level.
We clone your CRM tool or use a test instance. You give us 20-30 typical questions you expect. We test each. Success criterion: correct answers plus right permission application plus audit log consistent.
System goes live, audit log runs, we track tool call volume, success rate, permission violations (should be zero). Monthly monitoring report. When your tool's API changes (HubSpot deprecating v3 for v4, Shopify moving REST to GraphQL) we pull the connector along.
Pricing
Standard tools with existing MCP server (HubSpot, Salesforce, Notion, Linear): from 2,500 EUR setup plus 199 EUR/mo. Custom MCP for specialised tools (DATEV, Lexware, German ERPs): from 4,500 EUR setup plus 299 EUR/mo. Read-only connections are cheaper than bidirectional. LLM costs run separately on your OpenAI or Anthropic account.
See pricing and packagesFAQ
Standard stack: HubSpot, Salesforce, Pipedrive, Notion, Linear, Airtable, Microsoft Dynamics, Zoho — these have standard MCP servers we just configure and extend with a permission layer. Mid tier: Shopify, WooCommerce, Stripe, QuickBooks, Xero — custom MCP needed but standard APIs. German tools: DATEV, Lexware, sevDesk, BuchhaltungsButler — custom MCP, often with API limits we have to work around.
Three options: 1) RPA bridge (we build a bot that operates the tool like a human) — works but is fragile and expensive to maintain. 2) Database bridge if the tool has its own DB we can access read-only. 3) Workaround via export-import pipelines (e.g. CSV export every hour). We recommend option 2 if possible, otherwise pick a new tool.
MCP is the 2026 standard for LLM tool connections. Advantages: standard protocol, directly supported by Claude and ChatGPT and Cursor and Codex, built-in permission models. We use REST API bridge when you also need the tool for non-LLM use cases (classic web apps, mobile apps). Often it's both in combination.
Four layers: 1) OAuth or API key with secret manager (Vault, AWS Secrets Manager, no plain text) 2) Permission layer on MCP tool level with per-user roles 3) Rate limiting so no LLM loop overloads your CRM 4) Audit log for all calls. Plus: sensitive fields (salaries, HR data) are blocklisted in the MCP server code, the LLM never sees them.
Permissions are defined per user. Example: management sees all deals, field sales only deals in own territory, intern read-only on contacts without phone numbers. We map user roles from your existing tool (e.g. Salesforce roles) to MCP permission sets. When the tool changes, permissions update automatically.
Standard MCP servers (HubSpot, Salesforce) get updates from us or from the tool community — we pull updates and test. Custom MCP is maintained by the studio (included in 199-399 EUR/mo hosting). We track API deprecation notices from OpenAI, Anthropic, HubSpot, Salesforce — on breaking changes we've adapted the system before the old version is shut down.
Next step
We look at which tool you want to connect, which fields are relevant and which connection type makes sense. Honest assessment, no sales pitch.
