Claude Code

Claude Code is Anthropic’s CLI coding assistant. It needs two things from fremforge: HTTPS Git access to clone repos, and a fremforge PAT for issue + PR operations.

One-shot setup

# 1. Mint a PAT at frem.sh/-/settings/tokens with these scopes:
#      repo:read  repo:write  issues:read  issues:write
#    Copy the ffp_… string.

# 2. Tell Git Credential Manager about it so HTTPS clones don't prompt:
git config --global credential.https://frem.sh.helper "!f() { echo username=token; echo password=$FREMFORGE_PAT; }; f"
export FREMFORGE_PAT=ffp_xxxxxxxxxxxxxxxxxxxxxx

# 3. Add fremforge to Claude Code's MCP / API config (~/.claude.json):

{
  "mcpServers": {
    "fremforge": {
      "command": "npx",
      "args": ["-y", "@fremforge/mcp-server"],
      "env": {
        "FREMFORGE_BASE_URL": "https://frem.sh/_app/api/v1",
        "FREMFORGE_TOKEN": "ffp_xxxxxxxxxxxxxxxxxxxxxx"
      }
    }
  }
}

Or, if you’d rather Claude Code use the raw REST API without an MCP server, drop a CLAUDE.md at your repo root:

# fremforge config

REST API base: https://frem.sh/_app/api/v1
OpenAPI spec:  https://docs.frem.sh/api/openapi.yaml
Auth header:   Authorization: Bearer ffp_…

Claude Code reads CLAUDE.md as project context.

Verify

curl -sf https://frem.sh/_app/api/v1/user -H "authorization: Bearer $FREMFORGE_PAT" | jq .login

That should print your fremforge login. If it 401s, re-check the PAT scopes.

What you can do from Claude Code

Action	Required PAT scope
Read repo files, search across repos	`repo:read`
Create / update PRs, push branches	`repo:write`
Read / search / comment issues	`issues:read`, `issues:write`
Read audit log (operator-mode only)	`audit:read`
Mint or revoke own PATs	`tokens:manage`

Sovereignty note

When Claude Code reads your code and sends it to Anthropic’s API, that’s between you and Anthropic — fremforge is not in the path. The fremforge PAT is purely for Git + REST access; it never authorises Claude Code to call any AI vendor on fremforge’s behalf.

If you want the server-side AI features (PR review on every push, /ai/complete from workflows), that’s a separate setup at the AI integrations trust page.

Cursor