TL;DR
Plug in your MCP Server to let RecordsKeeper.AI call your internal tools safely.
Faster automations, richer answers: the AI can fetch live data and take approved actions.
Minutes to set up: point to your server, choose tools/scopes, test, and enable.
Why connect MCP?
Imagine asking the AI: “Summarize account ACME, open a priority ticket, and post the plan in #customer‑success.” With MCP:
The AI pulls context from your CRM and data warehouse.
It creates the ticket in your ITSM tool - with fields you specify.
It posts the update in Slack/Teams - automatically, traceably.
Result: fewer app‑hops, less manual stitching, and answers that are grounded in your systems - not guesswork.
What this integration unlocks
Real‑time data access: Query CRMs, ERPs, wikis, warehouses, and internal APIs via MCP tools.
Action execution: Create tickets, update records, trigger jobs - only where you’ve granted permission.
Context‑rich responses: The assistant explains what it did and why, with links and artifacts.
Workflow automation: Tie multi‑step tasks to one prompt or a scheduled run.
Backstage pass, safely: You control which tools are visible, which are read‑only, and which require confirmation.
Before & after
Before: Copy/paste between five systems. Separate tabs. Missed context.
After: One prompt. Verified tool calls. Logged actions. Consistent outcomes.
Prerequisites
A reachable MCP Server (staging recommended for first setup)
Credentials (API key/OAuth) with the minimum scopes you need
Admin rights in your RecordsKeeper.AI workspace
Setup (2–5 minutes)
Open your RecordsKeeper.AI → API & Integrations → MCP Servers.
Click Connect MCP Server.
Enter Server URL and a Display name (e.g., “MCP – Staging”).
Choose Auth method (API Key / OAuth / None) and add credentials.
Discover tools and select which to expose (mark sensitive ones as Confirm before run).
Click Test connection to validate reachability, auth, and tool discovery.
Save & Enable. Assign access to projects/teams that should use this server.
Pro tip: Start with read‑only tools, then enable write actions once you’re confident.
How it works
RecordsKeeper.AI reads the MCP server’s tool catalog.
You whitelist tools and (optionally) enforce confirmation for risky actions.
During a chat or automation, the assistant may invoke tools with your prompt’s parameters.
Logs capture who ran what, inputs/outputs, and response codes for audit.
Governance controls
Tool‑level allow/deny
Environment labels: Staging, UAT, Prod
Per‑team visibility
Rate limits, timeouts, and retries
Security & privacy
Least privilege: Connect with the smallest scope that works.
Hardened credentials: Encrypted at rest; never exposed in chat.
Transparent logs: Every tool call is recorded with inputs (sanitized) and results metadata.
Data handling: Tool outputs follow your workspace retention policy.
Tip: For production, create separate MCP servers per function (e.g., “CRM RO”, “ITSM RW”) to isolate credentials and scopes.
Troubleshooting
Can’t connect
Verify DNS/reachability from RecordsKeeper.AI; check firewall/IP allowlists.
Confirm TLS certs (valid chain, correct SAN).
Recheck auth method and scopes.
Tools not showing
Ensure your MCP server advertises tools per spec and your account can list them.
Refresh discovery after changing server config.
401/403 during calls
Token expired or scope insufficient. Rotate credentials and retest.
Slow or flaky calls
Raise timeout/retry in Advanced settings; rate‑limit bursty tools.
Inspect server performance/logging.
Accidental actions
Toggle Confirm before run for write tools; move risky tools to a separate server.
You’ll know it’s working when answers get sharper, tickets open themselves, and your dashboards stay fresh - without anyone chasing them.