Best AI Agent Skills for Django and FastAPI (2026)

Python web developers got skipped in the first wave of agent skills. The ecosystem filled up with JavaScript tooling: Next.js helpers, Drizzle patterns, Netlify deployers. Django and FastAPI developers were left pointing their agent at a generic Python skill and hoping it understood the ORM.

That changed over the last eight months. There's now a usable stack, and the interesting part is that Django and FastAPI need almost opposite things from an agent. Django's problem is that the framework already knows everything and the agent doesn't have access to it. FastAPI's problem is that the framework knows almost nothing and the agent has to be told.

Here's what's installable today, with star counts and security posture from our directory as of August 2026.

The stack at a glance

FastAPI: fastapi-mcp is the one that matters

claude mcp add fastapi-mcp -- npx -y tadata-org/fastapi_mcp

At nearly 12,000 stars it's the most adopted Python web skill in the directory, and the reason is a design decision that sounds small. It exposes your existing FastAPI endpoints as MCP tools with auth intact, rather than making you write a parallel tool definition for each route.

That distinction matters more than it reads. The usual way to give an agent access to your API is to hand-write a tool schema per endpoint, which immediately drifts from the real code the first time someone changes a Pydantic model. fastapi-mcp derives the tools from the app itself, so a route change propagates without a second edit. MIT licensed, last updated November 2025.

The auth handling is what I'd check first in a real deployment. Exposing endpoints to an agent means your agent inherits whatever permissions the token carries. If you wire it up with a service account that can delete records, an agent will eventually delete records. Scope the token to what the agent legitimately needs and test that the scoping holds before you point it at anything shared.

Django: the tooling is thinner and you should know that going in

claude mcp add django-mcp-server -- npx -y gts360/django-mcp-server

django-mcp-server is a Django extension that lets agents interact with your models, and at 368 stars it's an order of magnitude less battle-tested than the FastAPI equivalent. Our directory marks it unreviewed with the standard note: active community project, review the source before installing. Take that seriously here. It's a small enough codebase to read, and it's asking for access to your ORM.

I'd still install it, because the alternative is worse. Without model access, an agent working on a Django project is guessing at your schema from whatever it can grep out of models.py, and it will confidently write a query against a field that was renamed in a migration two years ago. Giving it real introspection removes an entire class of wrong answer.

What I wouldn't do is give it write access to a database with anything real in it. Point it at your local dev DB, let it read the schema, and keep migrations under human review. Django's migration system is the part of the framework where a plausible-looking mistake is most expensive to undo.

Pair it with mcp-server-mysql if you're on MySQL and want a second, deliberately read-only path to the data:

claude mcp add mcp-server-mysql -- npx -y benborla/mcp-server-mysql

Read-only is the entire feature. Verified trust tier, 2,026 stars, updated July 2026. There's also mysql-mcp-server from designcomputer at 1,355 stars if you want to compare. They solve the same problem with different opinions about connection handling.

The API-design layer

Both frameworks generate OpenAPI specs, and both generate specs too large for an agent to hold usefully in context once you're past about forty endpoints.

openapi-mcp-server exists for exactly that. It lets an agent query a complex spec in plain language instead of ingesting the whole document. 900 stars, MIT, updated July 2026, community trust tier. On a FastAPI project this is somewhat redundant with fastapi-mcp — you'd use it when consuming someone else's API rather than exposing your own. On Django REST Framework it's more clearly useful, since the DRF schema generation doesn't have an equivalent native bridge.

Azure's data-api-builder is the odd one on this list and worth knowing about even if you don't adopt it. It generates REST and GraphQL endpoints plus MCP tools directly from a database, which competes with the thing you'd normally write FastAPI to do. 1,491 stars, verified, updated August 2026. If you're standing up a straightforward CRUD layer over an existing schema, it's worth asking whether you need a Python service there at all.

Testing and discipline

The skills that improve Python web work the most aren't framework-specific.

claude skill add anthropics/skills/webapp-testing
claude skill add obra/superpowers/test-driven-development

webapp-testing is Anthropic's official Playwright skill, audited, and the only skill here carrying that status. It drives a real browser against your running app, which for a Django project means it can exercise the admin, the auth flow, and the templates rather than just hitting JSON endpoints.

sp-tdd from the Superpowers collection enforces red-green-refactor, and it's the skill I'd install first on any Python project where an agent is writing code. Agents are good at producing code that looks correct and bad at noticing they never ran it. Forcing a failing test before an implementation is the cheapest available guardrail.

Its sibling sp-systematic-debugging applies hypothesis-driven analysis instead of the agent's default behavior, which is to change three things at once and declare victory when the error message changes.

Grounding the agent in real docs

claude mcp add docs-mcp-server -- npx -y arabold/docs-mcp-server

Django's documentation is excellent and enormous, and a model's training data on it is always some months stale. docs-mcp-server indexes real documentation so the agent cites the version you're running rather than the version it remembers. 1,633 stars, verified, updated August 2026. This matters more on Django than on FastAPI simply because Django's API surface is larger and its deprecation cycle is longer, so there's more room for an agent to be confidently out of date.

Security notes worth reading before you install

Two of the skills here are unreviewed in our directory: django-mcp-server and openapi-mcp-server. That classification means active project, no formal audit, not that anything is wrong with them.

The practical checks, in the order I'd do them:

1. Read the install command. claude skill add pulls from a repo; claude mcp add ... npx -y executes a package. The second is a bigger ask. 2. Check what the skill needs access to. A database skill wanting write credentials in development is normal. The same skill wanting them in a .env that also holds production strings is not. 3. Look at the last-updated date. django-mcp-server was last touched in March 2026, the oldest date in this stack. 4. Run it against a scratch project first. Agents plus ORMs plus real data is the combination that produces stories.

Our guide on auditing a skill before installing it walks through the full checklist.

What I'd install today

For FastAPI: fastapi-mcp, webapp-testing, sp-tdd. That's the whole answer, and adding more before you've used those three is a way of avoiding the work.

For Django: django-mcp-server on a dev database only, docs-mcp-server, webapp-testing, sp-tdd. Read django-mcp-server's source first.

The honest summary is that FastAPI developers are well served right now and Django developers are not. Django's tooling gap isn't a technical problem — it's that the framework's conventions are strong enough that fewer people felt the pain that drives someone to build a skill. If you're a Django developer who's been thinking about publishing something, that's the open lane.

Developers working across other stacks can compare against our lists for Python, Go, and Laravel, or see our full list for developers.

FAQ

Do I need both fastapi-mcp and openapi-mcp-server?

Usually not. fastapi-mcp handles your own app. openapi-mcp-server is for consuming a third-party API with a large spec. Installing both makes sense when you're building a service that integrates with someone else's.

Is django-mcp-server safe to use in production?

I wouldn't. It's unreviewed with 368 stars and a March 2026 update. Use it against a development database, keep migrations under human review, and revisit when it has more adoption behind it.

Which skill helps most with Django migrations?

None of them directly, and that's deliberate advice rather than a gap in the list. Migrations are where agent mistakes are most expensive and hardest to reverse. Give the agent schema visibility through django-mcp-server so its suggestions are grounded, then write and review the migration yourself.

Can these skills work with Flask?

openapi-mcp-server, mcp-server-mysql, webapp-testing, and sp-tdd are framework-agnostic and work fine. The Django and FastAPI bridges don't. Flask has no equivalent skill in the directory yet.

What's the difference between claude skill add and claude mcp add?

claude skill add installs instructions the agent reads. claude mcp add runs a server that gives the agent new capabilities, usually with network or filesystem access. The second deserves more scrutiny before you approve it.