We're excited to announce MCP Management—a way for a workspace to curate the MCP servers Claude Code can use and inject them into every member's session automatically.
The Problem: Everyone Wires Up Their Own Tools
The Model Context Protocol is how Claude Code connects to the outside world—Linear, a database, your docs, internal APIs. It's powerful. It's also per-machine: each teammate edits their own .mcp.json, pastes in their own API keys, and hopes they got the args right.
The result is predictable:
- Half the team has Linear wired up; the other half doesn't.
- Someone commits a connection string into a config file.
- "Which args does the Postgres server take again?" gets asked in chat for the tenth time.
- A new hire spends their first afternoon copying setup snippets from a wiki.
Tools that should be a team capability end up as everyone's individual chore.
The Fix: Curate Once, Inject Everywhere
With MCP management, an admin defines a server once in the SuperClawd app. From then on, it's injected into every workspace member's Claude Code session on their next launch—correct args, correct env, no copy-pasting.
Two transports are supported:
- stdio — a local command (like
npx) with arguments and environment variables. - http — a remote server reached by URL, with request headers.
The Preset Catalog
The fastest way to add a server is the catalog. Click Add MCP, pick a preset, and the configuration is filled in for you. We're launching with:
- Linear — issues, projects, and docs. Signs in through your browser; no key to paste.
- Context7 — up-to-date, version-specific docs for the libraries you use.
Prefer something else? Choose Custom server and configure command / args / env (stdio) or url / headers (http) yourself.
Secrets, Handled Properly
MCP servers usually need a secret—an API key, a connection string, a bearer token. SuperClawd encrypts those values (args, environment variables, and headers) at rest, and only ever sends them—decrypted—to your authenticated CLI. Read-only members see secret values masked in the app.
To actually launch a server, Claude Code reads its config from disk, so the CLI writes the decrypted config into a per-launch folder on your machine that's cleaned up when the session ends. The principle is simple: we keep your secrets safe on our side, and never put our decryption keys on your machine.
Enabled, Experimental, and Favorites
Each server has two toggles that mirror the rest of SuperClawd, plus a favorite:
- Enabled — turn a server off without deleting it.
- Experimental — only inject it for members running in experimental mode, so you can trial a new server with a few people before rolling it out to everyone.
- Favorite — star the servers you reach for most so they sort to the top of the list. It's a personal quick-find marker and doesn't change what's injected.
Who Can Manage
Owners, admins, and maintainers can add, edit, enable/disable, favorite, and delete MCP servers. Members get the list read-only, with secret values masked. Server names are kebab-case (up to 30 characters), and each server takes up to 5 arguments and 5 environment variables. Workspaces can have 1 server on the free tier and 25 on paid.
Use Cases
- Onboarding: new hires get the team's tools on their first session—nothing to set up.
- Consistency: everyone connects to the same services, configured the same way.
- Security: secrets live encrypted in one place instead of scattered across laptops and config files.
- Rollout control: trial a new server in experimental mode before it reaches the whole team.
Get Started
MCP management is available now in every SuperClawd workspace—one server on the free tier, 25 on paid.
- Go to app.superclawd.com
- Open the MCPs page in your workspace
- Add a server from the catalog—or configure a custom one
- Your team's next Claude Code session picks it up automatically
Want the full details on transports, encryption, and injection? Read the docs at docs.superclawd.com.
Pro tip: Start with the one tool you wish everyone on the team had wired up. Add it once—now they all do.