# TuringCorp Decider **Category:** ๐Ÿ’ป Developer Tools **Repository:** https://github.com/TuringCorp-net/MCP **Views:** 0 **Installs:** 0 **Upvotes:** 0 **Directory Page:** https://allmcps.com/mcp/turingcorp-decider ## Description A judge for agent decisions: give it two defensible options, get the better one back. ## Claude Desktop Quick Installation Heuristic fallback โ€” verify the package name and runner against the repository README before running it. Uses `npx` (confidence: low): ```json "mcpServers": { "turingcorp-decider": { "command": "npx", "args": ["-y","turingcorp-decider"] } } ``` ## Documentation & README # TuringCorp โ€” MCP server A **decision model** for the calls that don't have a right answer. Your agent has two defensible options and has to pick one. Send both: a **panel of models** judges them together and returns the better one, **how far apart it judged them** (a calibrated confidence), and why. - **Endpoint (canonical):** `https://mcp.turingcorp.net/mcp` - **Aliases:** `https://mcp.turingcorp.net/` and `https://mcp.turingcorp.net/mcp/` - **Health:** `https://mcp.turingcorp.net/healthz` - **Transport:** Streamable HTTP, **stateless** (no session, no Durable Object). `GET /mcp` โ†’ 405 is normal. - **Protocol:** MCP `2026-07-28` (modern) **and** `2025-06-18` (legacy) โ€” one route serves both. - **Discovery methods** (`server/discover`, `tools/list`) need **no credentials** โ€” directory scanners can read the full tool list unauthenticated, by design. ## Making a decision: `decide` | | | |---|---| | **In** | `task` (string), `optionA` (string), `optionB` (string) โ€” all required | | **Out** | `job_id` (string), `betterOption` (`"option_A"` \| `"option_B"`), `confidence` (percentage string, e.g. `"76.7%"`), `reason` (string) | | **Annotations** | `readOnlyHint: true` ยท `openWorldHint: false` ยท `idempotentHint: false` | | **Timeout** | Reserve **180โ€“300 seconds** โ€” a decision is a long call. The timeout is a **client/host setting, not a tool parameter**: there is nothing to pass in the call. A 60s default cuts it off before the answer arrives; if that happens, do not call again โ€” retrieve by `job_id`. | | **Long calls** | A client that declares the `io.modelcontextprotocol/tasks` extension gets a task handle back instead of holding one connection open for minutes, and polls `tasks/get` โ€” see [Retrieving a result](#retrieving-a-result). Clients that do not declare it see no change at all. | **A successful call returns the decision inline.** `job_id`, `betterOption`, `confidence` and `reason` all arrive in the *same* tool result โ€” there is nothing to poll and nothing to fetch afterwards. The `job_id` is only for the case where the call never came back (timeout, dropped connection, client gave up waiting). `confidence` is **this service's own judgement of how far apart the two options were** โ€” a reference for your decision-making, not an instruction, not a result, and not a prediction of how the choice turns out. As a rough reading: โ‰ฅ90% clearly apart ยท 80โ€“90% apart, less clearly ยท 70โ€“80% a closer call ยท <70% close to evenly matched. These ranges are descriptive only. **Choose your own threshold for your own use case; for high-stakes or irreversible decisions apply your own review policy.** Observed accuracy for each range, and how it was measured, is published at โ€” that page is the authority on the bands and their measured accuracy. โš ๏ธ `idempotentHint: false` is an honest declaration: a retried call is a new call. **Record the `job_id` the result returns** โ€” a decision you have already paid for can be collected for **7 days** with the same credential (`GET https://api.turingcorp.net/v1/jobs?job_id=`; without the id, `GET https://api.turingcorp.net/v1/jobs` lists the job ids for that credential). Retrieve instead of retrying. ## When not to use it - **More than two options.** It compares exactly A and B โ€” there is no third slot, and it will not rank a list. - **Anything you can compute or verify.** A spec, a test, a price, a document: if something objective decides it, use that. This is for choices where no objective rule does. - **Factual questions.** It picks between two candidates; it does not look anything up. - **Speed-critical paths.** A decision is a long call and a paid one. Do not put it behind a request that has to answer in seconds. - **High-stakes irreversible calls without review.** Route on the confidence and keep your own review policy. ## How to fill the three arguments - `task` โ€” state the decision **neutrally**, without leaning toward either side: *"Which email do I send?"*, not *"Should I send the honest one?"* - `optionA` / `optionB` โ€” one **concrete** option each, plus the case for it. Plain text or Markdown, any length; keep the two sides roughly comparable so the comparison is fair. - **One option = one plan.** Do not bundle alternatives into a single side ("go indoors *or* postpone"): it compares the two slots, it does not split one of them for you. ```json { "task": "Which version of the delivery-slip email do I send to a client we want to keep?", "optionA": "Short and direct: the integration took longer than planned, delivery moves to the 24th, everything else is unchanged.", "optionB": "Warmer and longer: thank them for the kickoff, explain that dependencies took more time, offer to walk through the details." } ``` More patterns โ€” choosing the version you send, thresholding on the confidence, and writing the two options so the comparison is fair: **[docs/COOKBOOK.md](https://github.com/TuringCorp-net/MCP/blob/HEAD/docs/COOKBOOK.md)**. ## Retrieving a result Every call returns a `job_id`. **A call that is allowed to finish returns the decision inline โ€” the `job_id` is only for the case where it never came back.** โš ๏ธ **How to retrieve โ€” use `get_result`.** Retrieval needs the Agent Pass, and **inside an MCP host the agent usually does not have it** โ€” so retrieval is a **tool call**, not a REST call: `get_result` is authenticated by the host, which attaches the pass for you. **An agent never handles the pass**, and the credential is deliberately not a tool argument (it would end up in prompts and transcripts). - **`get_result` with a `job_id`** โ€” that job's status, and once it succeeded, the same decision body the original call returned. - **`get_result` with no argument** โ€” the job ids this credential created in the last 7 days. This is the case that matters most: when a call is cut off, you never received an id. - **Read-only and free** โ€” it starts no new work and costs nothing, so it is safe to retry (`readOnlyHint: true`, `idempotentHint: true`). - A host that declares the `io.modelcontextprotocol/tasks` extension can poll `tasks/get` instead. **Most clients do not declare it yet**, which is why `get_result` exists. **Operator/host, with the pass (the same data over REST):** - **With the id** โ€” `GET https://api.turingcorp.net/v1/jobs?job_id=` using the same Agent Pass. - **Without it** โ€” `GET https://api.turingcorp.net/v1/jobs` lists the job ids that credential created in the last 7 days; then fetch one as above. The list carries only the job id, the product and a timestamp. - **Nothing yet** โ€” an empty list is **not** an error, and there is **no** `job_id=0` placeholder: `{"object":"list","window_seconds":604800,"data":[]}`. Asking for `0` returns `404 No such job.` Retrieval returns the job's status and, once it succeeded, the same body the call itself would have returned. A job that is not yours, or older than **7 days**, is reported as unavailable. `GET https://api.turingcorp.net/v1/account` returns `{"account_id":"โ€ฆ"}`. **The practical answer is not to need retrieval: reserve 180โ€“300 seconds** so the call is allowed to finish. ## See it decide first There is no free tier, so we publish the evidence instead: **27 real decisions, recorded verbatim** โ€” the question, both options, which one was preferred, the confidence reported, and the reason. Nine domains, three cases each: Writing ยท Tech ยท Business ยท Research ยท Career ยท Money ยท People ยท Travel ยท Everyday. โ†’ **** โ€” dataset: [`examples.json`](https://github.com/TuringCorp-net/poe-demo-public/blob/main/examples.json) Read the confidence column first. On ordinary, closely matched questions it reports **70โ€“90%, not 99%** โ€” and that is the point of a calibrated number rather than a decorative one: a high value means the comparison was decisive and the pick can be acted on, a low value means the two options really are close and the choice stays yours. In this set the range is **27.3%โ€“88.3%, median 74.0%, none above 90%**, because these are everyday close calls, not easy ones. What the bands deliver when they *are* decisive is published at . ## Authentication Send an **Agent Pass**: `Authorization: Bearer `. Issued at **** โ€” self-service signup (email verification) โ†’ top up โ†’ get a pass. A pass is valid for **7 days** and can be re-rolled. If a call is refused as an invalid credential, sign in there again and re-roll. Errors fall into **two classes**, and they arrive in different places: > Credential problems are rejected before the call (HTTP 401 + `WWW-Authenticate`). Business failures come back as a tool result with `isError: true` plus a JSON block `{error, http_status, action_url, message}` โ€” `http_status` is the upstream status; the tool call itself is HTTP 200. | Class | What comes back | Examples | |---|---|---| | **Credential** | HTTP `401` + `WWW-Authenticate` โ€” no tool result is produced | no credential ยท expired or invalid pass (`invalid_credential`, with a re-login URL in the message) | | **Business** | HTTP `200`, tool result `isError: true`, plus a JSON block | insufficient balance (`http_status` `402`, `action_url` pointing at top-up) ยท over quota (`http_status` `429`) | Branch on the **HTTP status first**: a `401` will not succeed on retry without a new credential, whereas a business failure was a call that ran and was declined โ€” read the block for the reason and the next step. The detailed contract is under *Errors are machine-readable* below. Entry rate limit: **120 requests / 60 s / client IP**. This is flood damping, not a quota โ€” the real per-key quota is enforced separately. The counter is approximate: it is maintained per edge location and is eventually consistent, so a brief overshoot is possible. ## Two ways in โ€” and they are not the same - **Through an MCP client or host** (Claude Code, Cursor, VS Code, Codex, TRAE, Cozeโ€ฆ). The host holds the credential and attaches it for you: **you do not set an `Authorization` header yourself, and in most hosts you cannot**. The timeout is a host setting, not something you pass in the call. - **Directly against the REST API** (). Here you *do* send `Authorization: Bearer ` yourself, and you can retrieve a job by id. โš ๏ธ The part that catches people out: **being able to call the tool through a host does not mean you can reach the REST API.** The host may never hand you the underlying Agent Pass, so a job-retrieval call you make on your own can come back `401`. If your host declares the Tasks extension, retrieve through the tool surface instead (see [Retrieving a result](#retrieving-a-result)); if it does not, retrieval may simply not be reachable from where you are. ## Quickstart You need an **Agent Pass** to call the tool. Discovery (`tools/list`) works without one โ€” calling `decide` does not. ๐Ÿ”‘ **Where the pass lives matters.** If your client can keep it out of the file, do that โ€” VS Code's `${input:โ€ฆ}`, Codex's `bearer_token_env_var`. A pass written in plain text inside `mcp.json` or any client config is readable by **every agent and process that can read that file**, and assistants do read their own config โ€” one can print your pass straight into its output. Treat a client config as public within your machine. ### 1. Get an Agent Pass Self-service at ****: sign up with an email address, verify it, add credit. The pass is shown once; it is valid for **7 days** and can be re-rolled at any time. Agents can do the same thing over the API โ€” see that site's `/llms.txt`. ### 2. Point your client at the endpoint **Any client that reads a JSON config โ€” Cursor, Windsurf, Claude Desktop:** ```json { "mcpServers": { "TuringCorp": { "type": "http", "url": "https://mcp.turingcorp.net/mcp", "headers": { "Authorization": "Bearer " } } } } ``` Keep the `"type": "http"` line: a client that reads a `url` entry with no `type` treats it as a local stdio server and skips it. **Cline needs a different `type` โ€” `"streamableHttp"` (camelCase, no hyphen).** Anything else, or omitting it, makes Cline fall back to **SSE**, which this server does not speak, and the connection fails with **`405`**. The settings file is `cline_mcp_settings.json` (**MCP Servers** icon โ†’ **Configure** tab โ†’ **Configure MCP Servers**): ```json { "mcpServers": { "TuringCorp": { "type": "streamableHttp", "url": "https://mcp.turingcorp.net/mcp", "headers": { "Authorization": "Bearer " }, "disabled": false, "autoApprove": [], "timeout": 300 } } } ``` `timeout` is in **seconds** โ€” leave it at 300 or higher. Cline has had a bug ([#2296](https://github.com/cline/cline/issues/2296), closed 2025-06-23) where requests died after about a minute regardless of that setting; if a call is cut off at ~60 s, suspect that rather than the server, and **retrieve by `job_id` instead of retrying** (a retry is a second paid call). **A client that only speaks stdio** needs a bridge, and the credential must go in through `--header` โ€” `mcp-remote` does not read an `AUTHORIZATION` environment variable, so a config that only sets one connects, lists tools, and then fails on the first call with 401. The missing space after `Authorization:` is deliberate: some clients mangle spaces inside `args`. ```json { "mcpServers": { "TuringCorp": { "command": "npx", "args": ["-y", "mcp-remote", "https://mcp.turingcorp.net/mcp", "--header", "Authorization:${AGENT_PASS}"], "env": { "AGENT_PASS": "Bearer " } } } } ``` > โš ๏ธ A bare `npx -y mcp-remote@latest https://mcp.turingcorp.net/mcp` with no credential **will connect and list tools, then fail on the first call** with 401. That is expected โ€” discovery is open, execution is not. **Claude Code:** ```bash claude mcp add --transport http TuringCorp https://mcp.turingcorp.net/mcp \ --header "Authorization: Bearer " ``` ### 3. Sanity-check it yourself **Discovery โ€” no credentials needed:** ```bash curl -s -X POST https://mcp.turingcorp.net/mcp \ -H 'content-type: application/json' \ -H 'accept: application/json, text/event-stream' \ -H 'mcp-protocol-version: 2026-07-28' \ -H 'mcp-method: tools/list' \ --data '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{"_meta":{"io.modelcontextprotocol/protocolVersion":"2026-07-28","io.modelcontextprotocol/clientCapabilities":{}}}}' ``` **Calling the tool** additionally needs `mcp-name: decide` and the credential: ```bash curl -s -X POST https://mcp.turingcorp.net/mcp \ -H 'content-type: application/json' \ -H 'accept: application/json, text/event-stream' \ -H 'mcp-protocol-version: 2026-07-28' \ -H 'mcp-method: tools/call' \ -H 'mcp-name: decide' \ -H "authorization: Bearer $AGENT_PASS" \ --data '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"decide","arguments":{"task":"Pick a launch date","optionA":"Ship now","optionB":"Wait two weeks"},"_meta":{"io.modelcontextprotocol/protocolVersion":"2026-07-28","io.modelcontextprotocol/clientCapabilities":{}}}}' ``` ### Versions Two version numbers exist and they are **not** meant to match: | | meaning | |---|---| | **Registry `version`** (`server.json`) | the **listing** version - the metadata record (endpoint, auth declaration, icon, links). Advances when the listing changes. | | **`serverInfo.version`** (from the live endpoint) | the **runtime** version of the deployed server. Advances when behavior changes. | A listing-only change (for example a corrected link) advances the registry version without touching the runtime, so the two can legitimately differ. ### Protocol details worth knowing - **`Accept` must contain both** `application/json` **and** `text/event-stream`, or you get 406. - **Protocol versions:** one route serves both generations. - **Modern `2026-07-28`** โ€” no `initialize` handshake; use `server/discover` and `_meta`. - **Legacy** โ€” standard `initialize` handshake. Measured negotiation: a client asking for `2025-11-25`, `2025-06-18`, or `2025-03-26` is answered with **exactly the version it asked for**; a modern `2026-07-28` `initialize` (which is not part of that generation) is answered with `2025-11-25`. - Server capability discovery via `tools/list` and `server/discover` needs no credentials. - **Errors are machine-readable.** There are two distinct kinds, and they arrive differently โ€” so branch on the right thing: | What failed | How you see it | What to branch on | |---|---|---| | **Credential** missing or invalid | **HTTP `401`** with a `WWW-Authenticate` challenge (carrying `error="invalid_token"` and `resource_metadata`); the body is a standard JSON-RPC error with `error.code = -32001`. No tool result is produced. | the **HTTP status code** | | **Business** failure โ€” insufficient balance, quota, upstream error | a normal tool result (`isError: true`) whose **second `content` block** is a single JSON object you can `JSON.parse`, e.g. `{"error":"insufficient_balance","http_status":"402","action_url":"โ€ฆ","message":"โ€ฆ"}` | the **`error` field** inside that block | In short: credential problems are rejected before the tool runs, so they never appear as a tool result; everything else that fails after that does, with a machine-readable block attached. - **Authorization discovery:** `401` responses carry a `resource_metadata` pointing at `/.well-known/oauth-protected-resource`. Note that this server uses a **static bearer credential, not OAuth** โ€” that document says so explicitly rather than sending you into an OAuth flow that does not exist. ## Not for browser clients This endpoint deliberately does **not** serve browser-based (cross-origin) MCP clients: requests carrying a browser `Origin` header other than localhost are rejected with 403, and no CORS headers are returned. Browser tooling such as the MCP Inspector works by proxying through a **local** process, which sends no `Origin` โ€” that path works normally. If you need a browser page to reach this server, run a proxy you control rather than calling it cross-origin. ## What this is not - **No other tiers.** Other TuringCorp capabilities are not reachable through this endpoint. - **No SLA.** No availability commitment is offered, and none should be inferred. - **Not an autopilot.** Decider is a component you call. How you gate on it โ€” thresholds, human review, retries โ€” is your policy and stays yours. - **`decide` has no idempotency key โ€” record the job id instead.** A retried decision is a new paid call; collect the original with `get_result`. ## Use it from a coding agent: the Agent Skill There is an **[Agent Skill](https://github.com/TuringCorp-net/MCP/blob/HEAD/skills/decider/SKILL.md)** for coding agents (Claude Code, Codex, and anything that reads the same skill convention). It teaches an agent *when* a two-option judgement is the right tool, how to write the two options so the comparison is fair, and how to read the confidence without over-reading it. Install it by pointing your agent at this repository, or copy `skills/decider/` into your skills directory and add the MCP server to your client (see [Quickstart](#quickstart)). ## About this repository This repository carries the **discovery metadata** for the server above โ€” what directory services and clients read to find and describe it. - [`server.json`](https://github.com/TuringCorp-net/MCP/blob/HEAD/server.json) โ€” the official MCP Registry manifest - [`docs/DIRECTORIES.md`](https://github.com/TuringCorp-net/MCP/blob/HEAD/docs/DIRECTORIES.md) โ€” where this server is listed, and what each directory requires - [`docs/COOKBOOK.md`](https://github.com/TuringCorp-net/MCP/blob/HEAD/docs/COOKBOOK.md) โ€” worked patterns for handing `decide` the right kind of call - [`skills/decider/`](https://github.com/TuringCorp-net/MCP/blob/HEAD/skills/decider/SKILL.md) โ€” the Agent Skill This server is also listed on Smithery: **** The server itself is a hosted remote endpoint; nothing here needs to be installed or run. ## License See [LICENSE](https://github.com/TuringCorp-net/MCP/blob/HEAD/LICENSE).