Binding Doctor

Wrangler ships your code. Doctor wires your stack.

bdr is a CLI and an MCP server that reconciles Cloudflare bindings between your code, your wrangler.toml, and your account. It lets an AI agent (Claude Code, Codex, etc.) take a Worker from env.CHAT_DB to a deployed app without ever asking you to copy a database ID.

$ bdr diff
BINDING         KIND     DECL  REF  LIVE  STATUS
--------------  -------  ----  ---  ----  ----------------
CHAT_DB         unknown  —     yes  —     missing-declared
SESSIONS_KV     unknown  —     yes  —     missing-declared
UPLOADS_BUCKET  unknown  —     yes  —     missing-declared

$ bdr apply --yes
  create d1   CHAT_DB        → my-app-chat-db-8i6x   (9c48b532-…)
  create kv   SESSIONS_KV    → my-app-sessions-kv-3h0f (0230f531…)
  create r2   UPLOADS_BUCKET → my-app-uploads-bucket-s000

Wrote 3 bindings to wrangler.toml.
Migrations [CHAT_DB]: applied 0001_init.sql

Re-run = no-op. Delete a resource in the dashboard? bdr apply heals it.

---

What it does

1. Parses wrangler.toml / wrangler.jsonc for declared bindings (d1_databases, r2_buckets, kv_namespaces, queues, vectorize).
2. Scans src/**/*.{ts,tsx,js,mjs} for env.X references.
3. Lists live resources in your Cloudflare account.
4. Diffs the three sources of truth.
5. Applies the diff: creates missing resources, writes IDs back to your wrangler config (with .bak), runs pending migrations/*.sql against new D1 dbs.

Idempotent. One command. No manual ID copying.

Install

Zero-install (recommended)

npx -p binding-doctor bdr diff

bdr and bdr-mcp both resolve via npx from the npm registry.

Global install

npm install -g binding-doctor
bdr diff

Credentials

Create an API token at https://dash.cloudflare.com/profile/api-tokens with these scopes: D1:Edit · R2:Edit · Workers KV Storage:Edit · Queues:Edit · Vectorize:Edit · Account Settings:Read.

Drop into .env in your Worker project:

CLOUDFLARE_API_TOKEN=...
CLOUDFLARE_ACCOUNT_ID=...

CLI

bdr diff [dir]            # three-way binding diff
bdr plan [dir]            # preview actions
bdr apply [dir] --yes     # create + writeback + run migrations
bdr migrate [dir]         # run pending D1 migrations only

MCP — wire into Claude Code

Add to .claude/mcp.json (or your client's MCP config):

{
  "mcpServers": {
    "binding-doctor": {
      "command": "npx",
      "args": ["-y", "-p", "binding-doctor", "bdr-mcp"],
      "env": {
        "CLOUDFLARE_API_TOKEN": "...",
        "CLOUDFLARE_ACCOUNT_ID": "..."
      }
    }
  }
}

Or, if globally installed:

{
  "mcpServers": {
    "binding-doctor": {
      "command": "bdr-mcp",
      "env": {
        "CLOUDFLARE_API_TOKEN": "...",
        "CLOUDFLARE_ACCOUNT_ID": "..."
      }
    }
  }
}

Tools exposed: bdr_diff, bdr_plan, bdr_apply, bdr_migrate.

The agent's loop becomes: write code → bdr_apply → ship. Zero human ID copying.

Why local, not remote

The MCP runs on your machine because it reads your local wrangler.toml and scans your source tree — your code is the desired state. A hosted Worker can't see your filesystem. (See BLOG.md for the design rationale.)

Binding kind inference

When code references a binding not declared anywhere, kind is inferred from suffix:

Suffix matches	Kind
_DB, _D1, DB	D1
_BUCKET, _R2, _STORAGE	R2
_KV, _CACHE, _SESSIONS, _STATE	KV
_QUEUE, _JOBS, _EVENTS	Queue
_VEC, _VECTORIZE, _INDEX, _EMBEDDINGS	Vectorize

Unknown kinds emit a warning instead of silently picking wrong.

Demo

bash demo/run.sh

Why this exists

Cloudflare's Agents Week 2026 shipped most of the agentic stack: API MCP, Skills plugin, Stripe-funded auto-provisioning, Project Think. But when an agent generates a Worker that references env.CHAT_DB, the human still copy-pastes a D1 UUID. There is no git status for bindings; no reconciler.

Wrangler is the build tool. bdr is the reconciler. It treats your code as the desired state and your account as the actual state, and closes the gap idempotently.

License

MIT.