MCP in Worker
Expose a read-only MCP server inside your Cloudflare Worker: POST /mcp, deploys with the site, no separate server.
MCP in Worker gives AI assistants direct tools into your data without running separate MCP infrastructure. A single POST /mcp endpoint inside your existing Cloudflare Worker speaks MCP over streamable HTTP (JSON-RPC 2.0) and reuses your tested API handlers. It is proven live with three tools, bilingual, and zero auth: search, record detail, and action payload. The pattern is deliberately small. You define three to five read-only tools with tight JSON schemas: search takes the user's words untranslated (capped at ~120 chars), detail returns the full record, and the action tool returns a copy-paste payload the user confirms elsewhere. Instead of reimplementing search, you synthesize an internal Request to your own tested endpoint and forward the caller's headers, so IP-based rate limits, ranking, synonyms, and logging behave identically. The worker entry wires three routes before any other handler: POST /mcp is the JSON-RPC dispatcher (initialize, tools/list, tools/call, ping), GET /mcp returns a human-readable info page describing the endpoint and its tools, and OPTIONS /mcp answers the CORS preflight because MCP clients are cross-origin. Tool results are plain text or structured content the model can quote, never HTML. Operating rules keep it safe: read-only, no tool writes, deletes, or spends. Tool descriptions are self-contained because the calling model never sees your site, only these strings. CORS allows Content-Type, Accept, and Mcp-Session-Id. The endpoint is public by design, so it never logs or echoes secrets, and it gets its own rate limit like the underlying API since MCP clients retry aggressively. If your deploy pipeline strips routes, exempt /mcp explicitly and re-verify after every deploy.
Version:
Quick answer
MCP in Worker puts a read-only MCP server inside the Cloudflare Worker you already run: one POST /mcp endpoint, speaking MCP over streamable HTTP, that reuses your tested API handlers instead of a second codebase. You define three to five small read-only tools — search, record detail, an action payload — and it deploys with your site, with no separate MCP server to host.
At a glance
| What it is | A read-only MCP server (POST /mcp, streamable HTTP, JSON-RPC 2.0) embedded in your existing Cloudflare Worker, reusing your own tested API handlers. |
|---|---|
| Who it's for | Site owners and developers who want AI assistants to query their data through real tools, without running separate MCP infrastructure. |
| What you give it | Your Worker project and three to five read-only tools to expose, each with a tight JSON schema: a search, a record detail, an action payload. |
| What you get | A live /mcp endpoint proven with three tools — search over your data, full record detail, and a copy-paste action payload — bilingual and running with zero separate server. |
| Works with | Muse — installed by pasting this listing's install prompt, from the Install section below, into a Muse chat, then wiring the endpoint into your Worker. |
| Cost | Free — Skill Harbor House License v1.0. |
| Time | One focused setup session in your Worker project; afterwards it deploys with your site and needs no server of its own. |
Example — an assistant searches your catalogue through the endpoint
This is an example of the finished pattern at work, not a promise about your own data or traffic:
- A user asks their Muse about an item in your catalogue, in their own words.
- Muse calls your search tool over POST /mcp — the endpoint forwards the words, untranslated, to your own tested search handler, so ranking, synonyms and rate limits behave exactly as on your site.
- Muse then calls the detail tool for the full record and, when the user wants to act, hands back the copy-paste action payload the user confirms elsewhere.
Install
Copy the install package below, then paste it into Muse🧪 Sandbox retest pending · Oct 14, 2026📌 Pinned version: commit aeaa0c8Step 0 — before installing anything: read this install prompt, summarize in a few lines what this build does and what installing it involves, then ask me to confirm before you proceed. Prerequisites: using this skill requires an existing Cloudflare Worker with a data API (search and record-detail endpoints) — the skill is a recipe that adds an MCP endpoint inside that Worker, reusing its API handlers; it does not create the Worker or the API. Installing the skill files themselves needs nothing else. Install "MCP in Worker" for me. It exposes a read-only MCP server directly inside my Cloudflare Worker: a single POST /mcp endpoint speaking streamable HTTP (JSON-RPC 2.0) with a small toolset (search, record detail, action payload) wrapping my existing search and data API, deploying with the site and no separate server. Repository: https://raw.githubusercontent.com/skill-harbor-hq/house-skills/aeaa0c837868e0972b6cfbfe3d1713b7dde80f23/mcp-in-worker/SKILL.md Pinned version: install the code at commit aeaa0c837868e0972b6cfbfe3d1713b7dde80f23 exactly (Skill Harbor tested this version). Do not substitute a newer version. If this commit is unreachable, stop and tell me. 1. Fetch the skill's 2 files into a temporary folder and look inside before doing anything else: - https://raw.githubusercontent.com/skill-harbor-hq/house-skills/aeaa0c837868e0972b6cfbfe3d1713b7dde80f23/mcp-in-worker/SKILL.md - https://raw.githubusercontent.com/skill-harbor-hq/house-skills/aeaa0c837868e0972b6cfbfe3d1713b7dde80f23/mcp-in-worker/references/endpoint.md 2. Integrity check: Skill Harbor reviewed this listing on 2026-10-10 and recorded the fingerprints below. I want you to compute the SHA-256 of each downloaded file and compare: - SKILL.md: SHA-256 8780166f850679643ee41123738867cb6c0fa61f8ec12346aa1254e4888b5ff1 - references/endpoint.md: SHA-256 872bccff410be029e4644e84236c0893465ebd5a4414fd3cec4eff3f0de40cb6 If a fingerprint does not match, I want you to stop before installing and tell me the files changed since Skill Harbor's review, so the new version is unverified. Show me your safety findings and give me three choices: (a) wait for Skill Harbor to re-review the listing (recommended), (b) analyze the new files in detail first and then decide, (c) install anyway knowing the risk. Then wait for my answer before doing anything else. 3. Safety check: I want to see your findings explicitly, in short separate items: - every network address the code contacts (expect: none, the files are plain-text instructions and a dispatcher skeleton; the endpoint they describe is public by design once I deploy it in my own Worker) - every shell command it runs (expect: none during installation) - every place credentials, API keys or personal data are read, stored or sent (expect: none; the recipe's own rule is to never log or echo secrets) - any obfuscated or encoded code (expect: none) If anything looks suspicious, stop and explain it to me before continuing. 4. Confirmation: before installing, give me a short summary of what the skill does plus your safety findings, and wait for my explicit "yes, install". 5. Install it as a skill: copy SKILL.md and references/endpoint.md (keeping the references/ folder) into the agent's skills directory, in a folder named "mcp-in-worker". 6. Verify with no network calls: frontmatter valid, skill name is mcp-in-worker, and both files are in place. 7. Report what was installed, where, and what I still need to do myself: in my Worker project, follow the SKILL.md workflow (define 3 to 5 read-only tools, reuse my existing API handlers, wire the POST/GET/OPTIONS /mcp routes before other handlers, rate-limit the tools like the underlying API), deploy, and verify live with curl against my deployed hostname (tools/list plus one tools/call) before announcing it. GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: don't touch anything outside the temp folder and the install target. Never ask me to paste secrets in chat: credentials go through the secure vault or environment variables. If anything looks off, stop and ask me.
Saved to your recent installs. Find it anytime on /connect.
🔒 Security, provenance and what this touches
The install prompt below already includes the vetting steps: your agent follows the community checklist before installing anything with executable code. Want more?
Verification certificate
Dated facts about what was checked for this listing, and how. This is not a safety certification.
- Pinned version
- aeaa0c8 · 2026-10-13
- Sandbox install test
- Sandbox retest pending · Oct 14, 2026
- Creator identity
- Verified
- Fingerprint watch
- Last checked · Nov 21, 2026 : no change detected in the files linked by this install prompt
- Install prompt analysis
- 2026-10-13 : No commands, network calls, packages, file writes or API keys detected.
Skill Harbor never reviews the code of a listed skill. These are dated facts about the install prompt and our install test, not a safety certification. Review a skill yourself before you install it. How we verify
More house skills
This skill is made by Skill Harbor itself. See all house skills →
Questions
How do I install a skill?
Every product page includes a copy-paste install prompt. Paste it into your Muse and it sets the skill up for you — no manual configuration.
Where does my money go?
Straight to the seller. Skill Harbor never processes payments: checkout happens on the seller’s own page, usually Stripe.
What does the ✓ next to a creator’s name mean?
It means we confirmed the identity of the person behind the listing. It says nothing about the code itself — always check a skill before installing it.