FastEdge MCP Server - Development settings

End-users should follow STANDALONE-SETUP.md — prod image against prod endpoints, only GCORE_API_KEY required. This file is for in-house development.

---

In-house dev: prod schemas against preprod endpoints

The 99% in-house workflow: use the published image (prod schemas baked in) but route API calls at preprod. Schemas drift rarely, so prod schemas describing preprod endpoints work most of the time.

.vscode/mcp.json:

{
  "servers": {
    "fastedge-assistant": {
      "type": "stdio",
      "command": "bash",
      "args": [
        "-c",
        "docker run --rm -i --pull=always -v ${workspaceFolder}:/workspace -e WORKSPACE_ROOT=/workspace -e \"GCORE_API_KEY=$GCORE_API_KEY\" -e \"GCORE_API_BASE=$GCORE_API_BASE\" -e \"BATCH_MAX_CALLS=$BATCH_MAX_CALLS\" ghcr.io/g-core/fastedge-mcp-server:latest"
      ],
      "env": {
        "GCORE_API_KEY": "your_preprod_api_key_here",
        "GCORE_API_BASE": "https://api.preprod.world",
        "BATCH_MAX_CALLS": "5"
      }
    }
  }
}

Drop any env entry you don't need — unset vars fall through to the server's defaults (BATCH_MAX_CALLS → 5, GCORE_API_BASE → baked prod URL). The -e flags are harmless when the outer var is empty.

--pull=always

Keeps you on the current image (the fastedge-plugin release cycle drives version bumps here). Adds ~1-3s to MCP server startup for the manifest check against ghcr.io.

Offline or slow network? Drop --pull=always and manage upgrades manually:

docker pull ghcr.io/g-core/fastedge-mcp-server:latest

Environment Variables

Variable	Required	Default	Purpose
GCORE_API_KEY	Yes	—	API authentication (legacy FASTEDGE_API_KEY also accepted)
GCORE_API_BASE	No	Baked at build time (prod: api.gcore.com)	Runtime override for the Gcore API base URL. In-house devs set this to https://api.preprod.world to hit preprod with prod schemas.
BATCH_MAX_CALLS	No	5	Max calls per batch_execute invocation. Bump for batches that exceed 5 steps (total runtime still capped at 3 min).
WORKSPACE_ROOT	No	/workspace (Docker) / cwd (local)	Where the server looks for user files for build/upload operations. Usually left at the Docker default.

Building and running locally

pnpm install
pnpm run build
./build-local.sh

Schemas

API schemas are committed to src/generated/ and regenerated against prod on each plugin release. To regenerate manually:

pnpm run generate:schemas:prod

generate:schemas:prod defaults to SPEC_BASE_URL=https://api.gcore.com; you can still override it (e.g. SPEC_BASE_URL=https://… pnpm run generate:schemas:prod) if you need a different source.

Rebuild with preprod schemas (in-house only — edge case)

Only needed if you're actively testing an endpoint whose preprod/prod schema has diverged. For the 99% case, use GCORE_API_BASE=https://api.preprod.world above and keep the published image.

pnpm run build:preprod      # = generate:schemas:preprod + build:server
./build-local.sh            # builds fastedge-mcp-server:local
# point your .vscode/mcp.json image ref at fastedge-mcp-server:local

Tests

pnpm run test           # runs test:api + test:reference-index
pnpm run test:api       # node:test suite (API handlers, timeouts, batch budget)