--- title: "Agent Builder — Workflows that already know your business" description: "Drag-and-drop workflows that read your documents, follow your rules, and call your tools. Agents that already know your business." url: "https://certant.ai/agents" --- The Certant MCP server is included on Business plans and above. Connect an agent you already use, such as Claude Code, and it can set Certant up for you. See how it works # Agents that already know your business Drag-and-drop workflows that read your documents, follow your rules, and call the tools you already use. [Start free](https://certant.ai/register) [Book a demo](https://certant.ai/contact-sales) See what people build Put an approval step anywhere it matters. A contract-risk agent, read left to rightExample 1. **Your data is the first node.** Contracts, agreements and tickets. 2. **It classifies, then routes.** One step, a structured answer, branches that follow your rules. 3. **It calls your tools, not ours.** Salesforce, ServiceNow, Jira, GitHub, your own APIs. 4. **The right people, the right channel.** Slack, Teams, email, with the citation attached. The actual builder ## The builder as it ships Drag from the library, connect on the canvas, configure inline and save. ![Certant Agent Builder canvas](https://certant.ai/public/assets/screens/agent-builder-product-shot.png) The library, the canvas and the node inspector in one view. Library Thirty-one platform components in six colour-coded groups, plus nine connectors. Canvas Pan, zoom, branch and merge. Nodes and wires Every node declares what it takes in and what it gives back, so bad wiring is caught before you run. Configuration Every node's settings live on the node. Minimap A live overview of the whole flow. Click anywhere to jump there. Test Run the agent end to end against real or mocked inputs. Deploy Versioned releases, instant rollback, and a webhook or CLI hook for every agent. The difference ## Most agent builders reach for your data last Certant starts there. How it works ## Build it, code it, test it Build it visually Most of it is drag, connect, run. Drag and drop 31 components 9 connectors Drop in code where you need it Edge cases, anywhere in the flow. Python JavaScript Any node Test against real data Prove a step before it touches production. Replay a run Mock inputs Evaluate an AI step Workflow library ## What people build with it Pick a job and see the whole flow. Pick a job Every component ## Every component, one canvas Thirty-one platform components and nine connectors. **Plus 3,000 more** from the Certant MCP connector library. **Ask Certant to build it.** Describe the agent you want and the AI FDE builds it, then waits for your approval. [See the AI FDE](https://certant.ai/ai-fde) The detail ## Read the detail Open a heading to read that part. The longer version ## What it does, in full ### Agents that already know your business **Your data is the first node.** Contracts, agreements and tickets. The agent reads them before it acts. **It classifies, then routes.** "Is this contract risky?" One step, a structured answer, branches that follow your rules. **It calls your tools, not ours.** Salesforce, ServiceNow, Jira, GitHub, your own APIs. OAuth or a service account, your choice. **The right people, the right channel.** Slack, Teams, email, with the citation attached so the reviewer can see why. ### The builder as it ships Drag from the library. Connect on the canvas. Configure inline. Hit save. **Library.** Thirty-one platform components in six colour-coded groups, plus connectors for Xero, iMIS, SharePoint, Jira, Confluence, Obsidian, Notion, WordPress and Drupal. Drag any tile onto the canvas. **Canvas.** Pan, zoom, branch and merge. Auto-tidy straightens a tangled flow, and the minimap shows where you are in a large agent. **Nodes and wires.** Every node declares what it takes in and what it gives back, so bad wiring is caught before you run. Upstream values are referenced inline by node id. **Configuration.** No pile of modals. Every node's settings live on the node: schema picker, prompt editor, dialect, temperature, validation. **Test.** Run the agent end to end against real or mocked inputs. Every node shows its input, output and cost, and you can diff a run against the last one. **Deploy.** Versioned releases, instant rollback, and a webhook or CLI hook for every agent. Same builder, your perimeter. ### Most agent builders reach for your data last Every agent starts with everything you have connected: contracts, policies, records. So instead of "fetch this row, send that email", the agent can decide: _"is this contract risky? If it is, send it to the risk team and say why."_ Code when you need it. UI when you don't. ### What people build with it Pick a job and see the whole flow: every branch, every sub-tool, every integration. ### Every component, one canvas Thirty-one platform components and nine connectors. Click a tile to see what it does and when to use it. ### Ask Certant to build it The AI FDE sits in a drawer on every screen. Describe the agent you want in plain English and it builds it, then waits for your approval before anything goes live. Templates ## Templates to start from Every agent starts from a template or a blank canvas; the builder follows your system theme. ![Agent Builder template chooser, dark theme](https://certant.ai/public/assets/screens/agent-builder-templates-dark.png) he template chooser: 26 workflows available. What comes in, what the agent does, what goes out. Risk-scan incoming contracts New contract Score the risk, write why Risky ones escalated Transcripts into CRM notes Raw transcript Write a clean summary Note filed on the right account Escalate compliance breaches Breach found Open a ticket, attach the citation The right people notified Generate member onboarding packs One form Assemble pack and policy excerpts Welcome pack and calendar invite Triage support tickets New ticket Match against the knowledge base Auto-reply, or escalate when unsure For the CIO ## How it is governed Human-in-the-loop steps Pause the agent, ask a person, and resume exactly where it stopped. [Security](https://certant.ai/security) Structured inputs and outputs Every node takes JSON in and gives JSON back, so there are no surprise blobs to parse. [Architecture](https://certant.ai/architecture) Full execution logs Every run, every step and every prompt is recorded. [Changelog](https://certant.ai/changelog) Versioning and rollback Roll back to the last good version in one click. [Changelog](https://certant.ai/changelog) Runs on your infrastructure Same builder, your perimeter. [Security](https://certant.ai/security) MCP server ## Give your own agents what Certant knows Claude Code and other MCP clients can ask your knowledgebases, query your data and run your agents, under a key you issue. [Book a demo](https://certant.ai/contact-sales) [See which plans include it](https://certant.ai/pricing) Included on Business and Enterprise plans and every dedicated and sovereign deployment. Anything that changes live data waits for a person to approve it. The detail is under the headings below. Claude Code, connected to CertantExample Which of our projects mention cyanobacteria? Then run the portfolio agent and tell me what changed. `set_knowledgebase` `ask_knowledgebase` `run_agent` `get_agent_run` Reply Fourteen projects mention cyanobacteria, the latest in 2022. The portfolio agent ran and added two new reports to the research dashboard. Project 1190 · page 9 Agent run completed Each tool call runs under the key Claude Code was given. ### What your agent can do What it opens up ### Work from the agent you use Your agent connects to Certant and gets tools for each part of the platform. It sees one group of tools at a time and switches group when the job changes, which keeps the list short enough to use well. Ask your knowledgebases Questions come back answered from your documents, with the page each part came from. `ask_knowledgebase``search_knowledgebase``find_text` Query live data Strata turns the tables in your documents and feeds into data your agent can explore and query in plain English. Explore mode`query_data` Build and run agents Your agent can lay out a workflow with the Agent Builder's nodes, test it, save it and run it. `test_run``save_workflow``run_agent` Set watches on your data Monitors check your data for thresholds and gaps and start the action you chose when one trips. Monitors and actions mode Load and manage documents Upload documents, check how processing is going and run connector syncs on a schedule. `upload_document``get_ingest_status``run_sync_now` Check usage and credits Bring your usage and credit balance into the conversation before a big job starts. `get_usage``get_credit_balance` ### Keys and approvals Keys and approvals ### Decide what each agent may do Every connection uses a key that a person issued, with scopes that say what it may touch. Anything that changes live data waits for a person to approve it. **Scoped keys.** A key opens only the routes its scopes allow, and a read-only key can read and nothing more. You can revoke a key at any time. **Approval for live changes.** Draft changes go to a branch. Changes to live data, or anything with an effect outside Certant, wait in the approvals queue until a person signs them off. **Calls run as a person.** Each call runs as the user who owns the key and is metered to their account, so you can see who asked for what. ApprovalsExample Waiting for approval Turn on the monitor **"New report with no attachment"** and email the library team when it trips. **Asked by** Claude Code, using a developer's key **Scope** Strata draft **Effect** Sends email outside Certant ApproveDecline Nothing changes in live data until someone presses Approve. ### How the AI FDE fits in Built in ### The AI FDE uses the same tools The AI FDE is the assistant inside Certant. It calls the same tools as the MCP server, from inside the platform, under a short-lived key that drafts changes and sends live ones for approval. On a sovereign deployment the FDE runs on open models on your own GPUs, Qwen3.6 35B or GPT-OSS 120B. The assistant, its tools and your data all stay on your hardware, air-gapped if you need it. Its model list is limited to open models you can host yourself, which keeps its work inside your environment. [See the AI FDE](https://certant.ai/ai-fde) Two ways to reach the same tools Inside Certant, on your models AI FDE Certant tools Your knowledge and data Your own agent, over MCP Claude Code Certant MCP server Certant tools Both clients pass the same key, scope and approval checks. ### For your IT team For your IT team ### What your engineers will ask about | Setting | Detail | |---|---| | Clients | Claude Code, Hermes, OpenClaw, NemoClaw and any other MCP client. | | Transport | stdio, generally available, with one container per session. SSE in beta for remote agents. | | Direction | Inbound only. Your agent connects in, and the server holds no OpenAI or Anthropic credentials. | | Keys | Per-user keys, stored as a hash. Scopes cover workflow authoring, running and admin, read-only access, and Strata read, draft and commit. | | Tool groups | Twelve modes: six for Strata and agent building, and six for knowledge, chatbots, agent operations, data sources, dashboards and account. The agent sees the current mode's tools unless you list them all. | | Results | Every tool returns the same JSON shape: a summary, the data, and errors that each carry a hint for fixing the call. | | Versioning | The server image is pinned to your platform release. | | Extras | Workflow and Strata primers as MCP resources, and skill bundles that Claude Code loads directly. | ### Questions Questions ### Before you connect an agent Which plans include the MCP server? The Business and Enterprise plans, and every dedicated and sovereign deployment. The Free trial, Starter and Pro plans use Certant through its own interface. What does my agent's model provider see? Your agent's provider sees what your agent asks Certant and what comes back, as with any other tool the agent uses. Give the key only the scopes and knowledgebases the job needs. Can it all run inside our network? The MCP server and the AI FDE can. On a sovereign deployment both run inside your network, and the FDE uses your own models. An outside agent such as Claude Code still needs its own connection to its model provider. How do we connect Claude Code? Create a scoped key under Settings, then add the Certant server to Claude Code's MCP configuration. The server runs as a container pinned to your platform version. ### For developers: what a tool returns What a tool returnsExample ``` { "ok": true, "tool": "ask_knowledgebase", "summary": "Answered from 3 documents", "errors": [], "data": { "answer": "Three projects measured blooms…", "citations": [ { "document": "Project 1047 final report", "page": 31 } ] } } ``` ## Bring a job you keep doing by hand. Point an agent at it and watch what it reads before it acts. [Book a demo on your documents](https://certant.ai/contact-sales) [Start free](https://certant.ai/register) It runs in our cloud or inside your own walls. Nothing has to leave the building. Agent runExample Contract 2026-114 scored 4 of 5 for risk. It is with the risk team, and nothing has been sent to the vendor. Vendor SOW · clause 9.2 Waiting on approval