# Rulebase > Rulebase is AI agent monitoring for customer-facing agents. It monitors every conversation your AI agent handles, learns from what your human team resolved after a handoff, and shows whether the fix is a prompt, knowledge, tool or workflow change, so the agent resolves more. Built for AI product teams at financial-services companies and other regulated businesses. Rulebase also publishes an MCP server that lets Claude, ChatGPT or Gemini read the conversations held in helpdesks and contact centre platforms, plus an open-source library of 149 agent skills for customer-facing AI agents, customer operations, quality assurance, compliance and RevOps. Trusted by fast-growing and public financial-services companies including Qonto, Rho, and Lesaka. ## MCP servers for helpdesks and contact centres Most of these platforms have an MCP server, and more ship every month — some from the vendor, some commercial, many community wrappers on GitHub. They wrap the vendor API: fetch a ticket, search tickets, list contacts. That is integration, and it is useful. Rulebase is for the questions integration cannot answer — whether a bot is deflecting or merely containing, whether QA reviewers agree, whether a complaint was identified, whether first contact resolution is real. Those need the conversation rather than the recorded fields, joined across channels so one customer counts once, in enough volume to sample defensibly, with a stated method. Checked August 2026 against live search results. - [MCP directory — every MCP server for these platforms](https://rulebase.co/mcp): Vendor, commercial and community servers for the major helpdesks and contact centre platforms, and what each is good for. - [MCP for Zendesk](https://rulebase.co/mcp/zendesk): Swifteq ships a supported commercial server on the Zendesk Marketplace, plus community servers on GitHub. Zendesk's own MCP support is the client side, during a conversation. - [MCP for Freshdesk](https://rulebase.co/mcp/freshdesk): Freshworks ships an MCP gateway in early access for Enterprise, alongside several actively maintained community servers wrapping the Freshdesk API. - [MCP for Help Scout](https://rulebase.co/mcp/help-scout): Help Scout publishes its own MCP server, generally available to all customers, and community servers cover conversations, reports and Docs as well. - [MCP for Crisp](https://rulebase.co/mcp/crisp): Crisp published a demo MCP server on its own GitHub as a reference implementation, and two community servers exist. - [MCP for Aircall](https://rulebase.co/mcp/aircall): Community servers cover calls, contacts, SMS and AI insights with broad API coverage, plus hosted options from Zapier and viaSocket. - [MCP for Five9](https://rulebase.co/mcp/five9): Community servers reach the admin and supervisor APIs — agent states, active interactions, queue statistics — rather than the calls themselves. - [MCP for Genesys Cloud CX](https://rulebase.co/mcp/genesys-cloud-cx): A well-maintained community server wraps the Platform API with a one-click Claude Desktop bundle. Genesys' own MCP support is the client side, during the interaction. - [MCP for NICE CXone](https://rulebase.co/mcp/nice-cxone): CXone's MCP support connects out to external AI systems during an interaction — the client side. Reading the interactions and QA evaluations afterwards is a different job. - [MCP for Avaya](https://rulebase.co/mcp/avaya): Avaya Infinity orchestrates MCP live during a call — the client side. Reading the record afterwards is a different job. Platforms that already ship a first-party MCP server — Intercom, Front, Gorgias, Plain, Pylon, Kustomer and HubSpot — deliberately have no page here. Point your AI client straight at the vendor. ## Skills library - [Skills and Workflows Gallery](https://rulebase.co/skills): 149 agent skills for customer-facing AI agents, customer operations, quality assurance, compliance and RevOps. Free, open source and vendor-neutral — they read conversations from whichever helpdesk you already run. - Install them all with `npx rulebase-skills install --all`, or one at a time with `npx rulebase-skills install `. They work in Claude Code, Codex and Cursor. - Source: https://github.com/rulebase-co/rulebase-skills - Categories: CX operations, Quality assurance, Compliance, Data and integration, RevOps, and Rulebase workspace skills. - The skills run against any conversation export and do not require a Rulebase workspace. The MCP server does — that is where the helpdesk sync lives. - All but one skill is read-only. Anything that would change a record is marked as such and asks before it writes. ## Products - [Agent monitoring](https://rulebase.co/product/read-every-interaction): Monitor every conversation your customer-facing AI agent handles, connect each one to the customer outcome, and learn from human resolutions to prioritise fixes. - [Simulations](https://rulebase.co/product/simulations): Test agent changes against real failure scenarios and personas drawn from your own customer history, compare versions, and review regressions before release. - [Self improvements](https://rulebase.co/product/self-improvements): Turn recurring failures into proposed prompt, knowledge, and tool fixes. Review changes, test them, and measure customer outcomes. - [Insights](https://rulebase.co/product/insights): Find quality gaps, customer signals, and emerging risks across customer conversations. Also available: - [AutoQA — score against your SOPs](https://rulebase.co/product/score-against-your-sops): Reviews every human interaction in real time against your SOPs, replacing 3% sampling with 100% coverage. - [Coaching — coach automatically](https://rulebase.co/product/coach-automatically): Turns every QA finding into targeted, automatic per-agent coaching. - [Onboarding agent — clear onboarding stalls](https://rulebase.co/product/clear-onboarding-stalls): Catches each onboarding stall and runs follow-up until the account is transacting. - [Reactivation agent — reactivate dormant revenue](https://rulebase.co/product/reactivate-dormant-revenue): Finds dormant and never-transacted accounts and runs outreach until they transact again. - [Retention agent — catch churn before it leaves](https://rulebase.co/product/catch-churn-before-it-leaves): Flags closure risk while the customer is still engaged and acts to keep them. ## How it works - Monitor: Rulebase reads every conversation your AI agent handles, alongside the conversations your human team handles, across first-party and third-party support tools. - Detect: it flags conversations where the agent gave a wrong answer, took an unsafe action, or closed without resolving the problem. - Diagnose: for each handoff, it compares what the human did with what the agent could not do, and classifies the gap as missing data, missing knowledge, a missing tool or action, or a workflow problem. - Prioritise: gaps are ranked by how many human-handled conversations they cause, so the team fixes the ones that move resolution most. - Improve: each fix comes as a copy-ready prompt, knowledge or tool change, tested in simulation before release and measured on resolution afterwards. - The ground truth is what your own human team did and the customer accepted, not a static answer key. ## Who it's for - Heads of AI, AI product leads and conversational AI teams who own a customer-facing AI agent in production and are accountable for its resolution rate. - Financial-services companies (neobanks, payment processors, lenders, business banking) and other regulated businesses, where a wrong answer from the agent is a compliance risk as well as a cost. - Contact centre and customer operations leaders who need quality assurance across 100% of human interactions rather than a 3% sample. ## Company - [Product overview](https://rulebase.co/product) - [Media](https://rulebase.co/media) - [Careers](https://rulebase.co/careers) - [Contact](https://rulebase.co/contact) - [Trust & security](https://trust.rulebase.co/) ## Customer case studies - [Rho: 3x fewer escalations, 100% conversation coverage](https://rulebase.co/media/rho-case-study): Rho expanded review from a 3–5% sample to every customer conversation, cut escalations by 3x, caught 4x more issues, lifted containment by 10 percentage points (Rulebase customer data), and reduced manual review by ~90%. Includes customer interviews and Rulebase’s human + AI loop: learn from human resolutions, test bot fixes in simulation, and monitor outcomes. - [Kuda: from a 3% sample to 98% conversation coverage](https://rulebase.co/media/kuda-case-study-2b1cf95ab4): Kuda reviews 98% of roughly 350,000 monthly customer conversations, has routed more than 20,000 high-risk interactions, and saves over 50 hours of manual work each month. ## News - [Rulebase and Rho sign multi-year deal to automate customer operations](https://rulebase.co/media/rulebase-rho-multi-year-deal): Rho signed a multi-year renewal with Rulebase, scaling QA from a 3–5% manual sample to 100% of customer interactions, catching 4x more issues, cutting escalations by 3x, and reducing manual review by ~90% without adding headcount. - [Rulebase launches AI Revenue Agents for financial services](https://rulebase.co/media/rulebase-launches-ai-revenue-agents): Launch of the Rulebase Revenue Agent — durable per-account tasks owned by an agent until the customer converts.