codemode-workers
Expose any API to an LLM agent as two sandboxed MCP tools (search + execute) on Cloudflare Workers.
README
codemode-workers
Expose any API to an LLM agent as two sandboxed MCP tools (search + execute) on Cloudflare Workers.
Instead of registering thousands of tool schemas, you register two tools (search + execute). The catalog stays on the server; the agent writes JavaScript that runs in a fresh, sandboxed isolate. Same pattern as cloudflare-mcp, packaged for any API.
Docs: codemode-workers.pages.dev
Why
One tool per endpoint dumps every schema into the model's context. Code mode keeps the spec server-side and sends two tools, so the context cost stays flat however big the API is. Against the Urantia Papers API (58 endpoints):
| Approach | Tools | Tokens |
|---|---|---|
| One tool per endpoint, full | 58 | 3,489 |
| Code mode | 2 | 184 |
Run bun run eval:tokens <specUrl> to measure it for your own API.
Install
npm install codemode-workers
Bring your own MCP server SDK (e.g. @modelcontextprotocol/server). The library only needs a registerTool(name, config, cb).
Usage
// wrangler.jsonc
{
"worker_loaders": [{ "binding": "LOADER" }],
"services": [
{
"binding": "GATE_SELF",
"service": "<your-worker>",
"entrypoint": "Gate",
},
],
}
import { exports } from 'cloudflare:workers'
import {
createGate,
processSpec,
registerCodemodeTools,
} from 'codemode-workers'
export const Gate = createGate({ allowedHosts: ['api.example.com'] })
registerCodemodeTools(server, {
loader: env.LOADER,
catalog: {
get: async () => processSpec(await (await fetch(SPEC_URL)).json()),
},
api: {
baseUrl: 'https://api.example.com/v1',
outbound: () =>
exports.Gate({
props: { headers: { Authorization: `Bearer ${env.API_TOKEN}` } },
}),
},
})
search runs code over your catalog with no network. execute calls the API through the gate, which allowlists your host and injects the credential that the agent never sees. Full walkthrough and API reference in the docs.
Security
The credential lives in the gate, outside the sandbox, so agent code cannot read the token or reach any host but yours. search has no network. Every call is a fresh isolate. Details and residual risks: codemode-workers.pages.dev/security.
Develop
bun install
bun run check # typecheck + tests (workerd via @cloudflare/vitest-pool-workers)
License
MIT. See LICENSE.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
Neon Database
MCP server for interacting with Neon Management API and databases
E2B
Using MCP to run code via e2b.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.