Zelda MCP: Connect Stell Requirements, Specs, and Engineering Data to Your AI Tools

Zelda, Stell’s system engineering AI Agent, already helps your team draft requirements, verify compliance, and navigate specs from inside the product.

MCP, short for Model Context Protocol, is an open standard that lets AI tools connect to outside systems and pull real data or take real actions. Instead of copying information between apps, your AI assistant can just ask for it directly.

Starting today, you can connect Zelda to Claude Code and custom agents. The same tools Zelda already uses inside Stell are now reachable from wherever your team works.

Usecase

An engineer is assigned a new requirement on a flight control software update: the system must detect a sensor dropout and switch to a backup source within 200 milliseconds. Getting from that requirement to working code means finding the right spec in Stell, reading the requirement, its rationale, and related linked requirements, then bringing that into their editor to start writing tests and code. On a program with hundreds of requirements moving through review at once, that context switch happens dozens of times a day.

With Zelda MCP, the engineer stays inside their agent. They ask for the requirement by ID or name. Zelda pulls the exact text, the 200-millisecond threshold, and the linked acceptance criteria straight from the spec and hands it to the client. From there, the engineer writes a test that fails until the failover happens inside that window, then writes the code to make it pass. The test is traceable back to the exact requirement it was written against. 

Multiply that across a sprint. Fewer transcription errors between spec and code, less time lost switching tools, and a clearer line from requirement to test to implementation, the kind of traceability a program depends on for review.

This works for more than requirements. A few examples:

• Specs and requirements. Pull a specification's table of contents, a single section, or a full requirement with its links and comments.

• Test procedures. Look up a procedure family, a specific version, and its ordered steps, so a testing tool knows exactly what to run.

• Evidence and exceptions. Retrieve evidence records tied to a requirement, or check whether a waiver or deviation already exists before flagging something as a gap.

• Risk registers. Pull risks out of a hazard log by severity or likelihood for a report or dashboard.

• Programs and products. Look up which programs a product line has been instantiated into, without opening Stell to check.

If Zelda can already surface it inside Stell, an MCP client can now ask for it directly. Some of these tools can write back to Stell as well, and are always scoped to what that user's key allows.

Security and compliance

Every MCP key is tied to a specific user on Stell, not a generic service account. When that person's key calls a Stell tool, it runs with their permissions, the same ones that govern what they can see and do inside the Stell UI. Nothing gets more access through MCP than that person already has.

Keys come in three tiers: read, write, and write with delete. Most engineers pulling requirements only need a read key, so a customer can choose to hand out read access and stop there.

Keys are also short-lived by design, with automatic expiry and re-authentication built in rather than left to someone to remember. Revoking a key, or offboarding a user, cuts off access immediately.

Because MCP calls run through the same Stell API and the same permissions model as everything else, every action shows up in the MCP audit trail. If a key edited a requirement or moved a specification, that action is traceable in the UI, tied to the user who made it, the same as if they'd clicked through the product by hand.

MCP access is available now. Read the documentation in the app to get started.