Flipt MCP Server
Official# Flipt MCP Servers
[](https://github.com/flipt-io/mcp-server-flipt/actions/workflows/ci.yml)
[](https://www.npmjs.com/package/@flipt-io/mcp-server-flipt)
[](https://www.npmjs.com/package/@flipt-io/mcp-server-flipt-v2)
[Model Context Protocol (MCP)](https://modelcontextprotocol.io/) servers for [Flipt](https://flipt.io), allowing AI assistants to interact with feature flags.
This repository contains two packages, one per major Flipt version:
| Package | Flipt version | npm | Docker |
| --- | --- | --- | --- |
| [`packages/mcp-server-flipt`](packages/mcp-server-flipt) | [Flipt v1](https://docs.flipt.io/v1) | `@flipt-io/mcp-server-flipt` | `ghcr.io/flipt-io/mcp-server-flipt` |
| [`packages/mcp-server-flipt-v2`](packages/mcp-server-flipt-v2) | [Flipt v2](https://docs.flipt.io/v2) | `@flipt-io/mcp-server-flipt-v2` | `ghcr.io/flipt-io/mcp-server-flipt-v2` |
Not sure which one you need? If your Flipt server has environments and Git-backed storage, it is v2 — use `@flipt-io/mcp-server-flipt-v2`. See each package's README for installation, configuration, and the full tool catalog.
## Quick start
Flipt v2:
```bash
npx -y @flipt-io/mcp-server-flipt-v2
```
Flipt v1:
```bash
npx -y @flipt-io/mcp-server-flipt
```
## Development
This is a [pnpm](https://pnpm.io) workspace.
```bash
pnpm install # install all dependencies
pnpm -r build # build both packages
pnpm -r test # test both packages
pnpm -r check # typecheck both packages
pnpm lint # lint both packages
pnpm fmt # format both packages
# scope a command to one package
pnpm --filter @flipt-io/mcp-server-flipt-v2 dev
```
## Releasing
- v1: push a `vX.Y.Z` tag (unchanged legacy scheme)
- v2: push a `mcp-server-flipt-v2@X.Y.Z` tag
Each tag publishes the matching package to npm and its image to ghcr.io via the [publish workflow](.github/workflows/publish.yml). Bump versions with `pnpm version <bump> --no-git-tag-version` inside the package directory (the `version` lifecycle script keeps `src/version.ts` in sync), then create the tag manually using the scheme above.
## License
Apache 2.0
TDQS
Scored across 28 tools
Every tool has a clearly distinct purpose based on its name, targeting specific resources (flag, namespace, segment, constraint, etc.) and actions (create, delete, evaluate, get, list, update, toggle). There is no ambiguity or overlap between tools, as each combines a unique action with a specific resource type.
All tools follow a consistent verb_noun pattern with snake_case throughout, such as create_flag, delete_segment, evaluate_batch, list_namespaces, and update_flag. There are no deviations in naming conventions, making the set highly predictable and readable.
With 28 tools, the count is borderline high for a feature flag management server, suggesting it might be heavy but still manageable. It covers many operations, but the high number could overwhelm agents, placing it in the 16-25 range that feels heavy yet not extreme.
The tool set provides complete CRUD/lifecycle coverage for the feature flag domain, including create, get, list, update, delete, and evaluate operations for flags, namespaces, segments, and related resources. There are no obvious gaps, ensuring agents can handle all core workflows without dead ends.