Accelerate your modelling — and keep it updated
Every enterprise-architecture tool dies the same way: the model rots faster than a handful of architects can update it. Dependra structurally answers that — let AI coding agents do the hard work of taking inventory of your IT landscape and generate light-weight manifests in your source code repositories. Maintain them either manually or using AI coding agents and push via your CI pipeline to move your estate at the speed of your codebase, not your documentation backlog.
Fed by your agents, your CI, and your monitoring
Point your favourite coding agent such as Claude Code at an existing repo and let it do a deep AI audit of it, and push the real architecture — components, dependencies, datasets — straight into Dependra over our first-class MCP connector. Or have it generate a structured manifest, commit it to source control alongside your code, and let your CI/CD publish it automatically through a scoped, least-privilege API key.
The same MCP tools let your agent ask the model — "what's the blast radius if the payments queue dies?" or "What components are required to reliably deliver the customer experience capability?" — so the answer is an API call, not a two-day workshop.
Stream live health signals from your existing monitoring platform to Dependra and it recomputes impact, opens incidents and fires webhooks the moment something degrades. You can also publish your live landscape to Confluence and keep it in sync automatically. Your landscape maintains itself instead of rotting in a wiki nobody trusts.
And for the humans: the AI Copilot
Agents feed the model; people question it. The built-in AI Copilot is a chat drawer inside Dependra that answers in plain language — "what breaks if the payments queue dies?", "which capabilities depend on anything end-of-life?" — using the same grant-aware tools, so it only ever sees what the person asking is allowed to see. Bring your own LLM: your AI router, your perimeter, your rules. Writes go through an approval gate — the Copilot proposes, you apply.
MCP server & tools
A tenant-scoped Model Context Protocol server hosted in-process at /mcp. Query tools (impact, blast radius, search) and mutating tools (upsert application/component/dataset, add dependency, set health) — thin adapters over the mediator, honouring scopes and revenue redaction.
Scoped API keys
Least-privilege keys scoped read / write / health, carrying a tenant claim that scopes every call. Full lifecycle: create, rotate, revoke. Governed by scopes, not membership RBAC.
Manifest import
Paste or upload a YAML manifest — AI-generated from existing docs — preview the diff, and selectively apply it, with per-row permission evaluation, a public schema + example, and the dependra CLI validator. On every plan.
Health signal pipeline
Push component health at scale to a durable queue; a consumer recomputes the cached impact snapshot, runs the incident lifecycle and triggers webhooks. Sustained ingest is capped per plan.
Outbound webhooks Enterprise
Deliver the impact snapshot — cost, SLA and regulatory context — into your ITSM / alerting stack, with test-fire, delivery history and secret rotation. Idempotent delivery.
Tenant export
Your entire authored landscape as open, line-delimited JSON (NDJSON), gzip-compressed on request — walk away any time. The anti-lock-in guarantee, on every plan.
Ask your architecture — the model is queryable
Because the MCP tools are thin adapters over the same engine the product runs on, an agent can query the model as easily as it can update it: current impact, blast radius for any entity, unified search, list and read components and applications. The technical champion trusts a working quickstart more than a testimonial — so the MCP quickstart and the API examples are live developer proof, not slideware.
What Dependra is not
- Not a CMDB. We don't track every config item by hand. We model dependencies and compute blast radius — and the model maintains itself via agents and CI.
- Not a discovery agent on your network. You choose what to publish, through a scoped key or a previewed manifest. A pasted manifest always previews a diff before it writes; from CI, your pull request is the gate.
Point an agent at a repo and watch it populate
Start your trial, connect the MCP server or paste a manifest, and let your coding agent push the real architecture in. No payment required to trial.
Start free trial