Noodle Seed
Noodle Seed is agent connectivity infrastructure for software teams. It turns an existing product workflow into a governed MCP capability that ChatGPT, Claude, Copilot and Gemini can call, handling customer identity, permissions, secrets and audit. Currently in public beta.
What is Noodle Seed?
Noodle Seed is agent connectivity infrastructure: it operates the governed connection between a software product and the AI agents its customers already use. The premise stated across the site is that a workflow is easy to demo and hard to run in production, because identity, permission boundaries, credentials, reliability and audit have to be rebuilt for every agent surface. Noodle Seed supplies that shared layer while the vendor's own backend remains the system of record.
The developer path starts with a single TypeScript file. You author a server.ts declaring connectors, tools with Zod input and output schemas, resources, prompts and optional React views. A compiler resolves it into a versioned product definition, and three CLI commands carry it forward: noodle validate compiles the product surface and catches contract problems, noodle test proves the behaviour locally, and noodle deploy publishes the approved definition to the managed boundary. Authoring and testing need no account; signing in is required only to deploy. The documentation calls the design thin language, fat runtime, meaning the protocol wire, schema validation, connector execution and credential brokering live in one shared multi-tenant engine instead of being rebuilt in every project.
Around each call the runtime applies resource-bound OAuth, tenant routing, per-tool scopes and roles, scoped credential exchange, secret isolation and constrained connector egress, then emits audit events stripped of request bodies, tokens and signed URLs. Operators get metrics, events, session replay, filtered logs, webhook alerts, release history and deployment inspection. Host coverage is deliberately uneven and stated as such: protocol compatibility for ChatGPT, Copilot, Gemini and generic MCP clients, verified tools for Claude and Codex.
A second product line serves non-technical teams under the same brand. You enter a website address, the platform extracts the business and generates a ChatGPT app, a Claude-oriented experience and Halo, a site assistant that answers questions, captures leads into HubSpot or Salesforce, books through Google Calendar or Outlook, and sells from a Shopify catalogue inside the conversation. Around forty customer apps are listed and clickable in the ChatGPT app directory. The platform is in public beta and its documentation labels each capability as shipped, partial or roadmap.
What it does
- Expose an existing product workflow as a governed MCP capability that ChatGPT, Claude, Copilot, Gemini and any MCP client can call
- Declare an entire product surface in one TypeScript file: connectors, typed tools, resources, prompts and React views
- Validate, test and prove the whole thing locally before creating any hosted dependency, with no account needed
- Enforce customer identity, tenant routing, per-tool scopes and roles on every agent-facing action
- Broker service credentials at call time so secrets never reach specs, logs, widgets or agent context
- Operate the result with metrics, events, session replay, filtered logs, release history and rollback
- Turn a website into a ChatGPT app and a Halo site assistant with no code, for non-technical teams
When to use Noodle Seed / When not to
A quick filter to help you decide if Noodle Seed is the right fit.
When to use Noodle Seed
- SaaS vendors with a customer workflow they already monetise and want to expose to AI agents without rebuilding their backend
- Engineering teams that already run an MCP server and now need customer identity, tenant routing, scoped credentials and audit around it
- Product and platform leaders who want to prove demand on one named workflow before committing to a broad agent roadmap
- Agencies and consultancies building and hosting MCP apps for several clients on one runtime
- Non-technical small businesses that want a presence inside ChatGPT and Claude, plus a site assistant that books, qualifies and sells
When not to use Noodle Seed
- Anyone looking for a chatbot or an AI assistant to use directly: this is plumbing between a product and the agents, not a model you talk to
- Teams with no API, no MCP surface and no existing workflow, since there is nothing to connect
- Organisations that need SSO, SCIM, compliance exports or group-level governance today, as the site places all of these in private preview or later
- Regulated buyers who need a signed DPA, documented retention periods and a named hosting jurisdiction before signing, none of which the site publishes
- Anyone who needs mobile apps or a non-English interface, neither of which exists
How to use Noodle Seed
A typical end-to-end flow, from setup to results.
- Decide which single customer workflow to expose first, ideally one that customers already pay for or have explicitly asked for
- Install the Noodle Seed plugin in your coding agent, with a marketplace command for Claude Code and install commands for Codex and Cursor
- Read the quickstart in the developer documentation to understand tools, resources, prompts, connectors and widgets
- Author a server.ts that declares your HTTP connector, its base URL, allowed origins and API credential held as a named secret
- Declare each tool with its description, required scopes, allowed roles, Zod input and output schemas and its fulfil function
- Attach a React view to any tool that deserves an in-host interface
- Run noodle validate to compile the product surface and catch contract problems
- Run noodle test to prove the behaviour locally, without an account and without creating a hosted dependency
- Sign in and run noodle deploy to publish to the managed runtime, then connect the endpoint from ChatGPT, Claude, Codex or any MCP client
- Watch metrics, events and filtered logs from the console, and roll back to a stored deployment if a release misbehaves
Pros & Cons
Pros
- One typed definition drives the protocol surface, the in-host experience, the client connection and the operational surface at once
- Complete local proof before any hosted dependency, and no account required until you deploy
- Your backend stays authoritative, so there is no migration and no duplication of business logic
- Identity, tenancy, per-tool authorization and credential brokering come with the runtime instead of being rebuilt per project
- Unusually candid security page that separates what ships today from what sits in private preview
- Priced per billing account rather than per seat, with pooled quotas and published overage terms
- Permanent free plan with no card, one million pooled MCP calls and one production app
Cons
- Public beta on the platform and Alpha in the documentation: limits may change and the terms carry no warranty of any kind
- Much of the governance story is announced as private preview, including SSO, SCIM, compliance exports and group-level policy
- Privacy policy is an explicit beta summary with no GDPR section, no retention periods and no data processing agreement
- No hosting country or region is declared, which sits awkwardly next to private connectivity being a selling point
- The published subprocessor list names only two providers and the site says it is incomplete
- No postal address, no phone number, and a single email address covering legal, privacy and support
- Two product lines with two unreconciled price ladders and different quota units, plus implementation fees on top
Pricing & Plans
A permanent free plan is available on both product lines, requires no payment card and includes one million pooled MCP calls and one production app. The cheapest paid entry point is the Pro plan of the developer platform at USD 30 per month, billed per billing account before applicable taxes. The no-code line starts higher, at USD 50 per month. Annual billing grants two months free on the platform plans and a 17 percent reduction on the no-code plans. Overage is charged at USD 5 per started block of one million additional calls. Implementation packages are priced separately, from USD 1,500 to USD 25,000 and above.
- 1M pooled MCP calls per month
- 1 production app
- local development without an account
- hosted deployment
- no card required
- 10M pooled MCP calls
- up to 5 production apps
- team access
- managed secrets
- deployment history and rollback
- email support
- 100M pooled MCP calls
- up to 25 production apps
- custom domains
- policies and audit logs
- release history and rollback
- priority support
- custom usage and app limits
- SSO
- SCIM and audit exports
- private connectivity
- dedicated deployment options
- contracted SLA and implementation support
- 1 user
- 100 AI responses per month
- one free AI discoverability audit
- docs and community support
- 1 user
- 1
- 000 AI responses per month
- monthly discoverability audits
- chat support with a 48-hour SLA
- up to 10 users
- 10
- 000 AI responses per month
- daily audits and automated testing
- Claude workflows and Halo deployment
- Shopify and custom system integrations
- selective programme for teams whose requirements go beyond Pro and who co-build with the engineering team
- Discovery Sprint USD 1
- 500 over 3 days
- Production Launch USD 5
- 000 over 5 days
- Branded Experience USD 8
- 000 over 10 days
- Multi-Workflow Bundle USD 15
- 000 over 15 days
Data, GDPR & hosting
A consolidated view of how Noodle Seed handles your data.
GDPR overview
Compliance is claimed but not documented. Every page of the main site carries a GDPR badge in the footer, next to a SOC II badge, and the structured data of the no-code product page lists the tool as SOC 2 Type 1 certified and GDPR compliant. Nothing behind those claims is published. The privacy policy contains no GDPR section at all: no legal basis, no retention period, no list of data subject rights, no data protection officer, no Article 27 representative and no mention of international transfers. It presents itself as a summary for the public beta, to be expanded into a full policy before general availability. Access and deletion requests are handled by writing to the general contact address. No data processing agreement is published or offered.
Who owns the data?
Ownership stays split along a clear line. The terms make the customer responsible for the servers they author and deploy, for the data those servers process and for any credentials they configure, and the marketing pages repeat that the customer's own product remains the system of record for data, workflows, permissions and business rules. Noodle Seed keeps only what running the service requires: account identity received through Google sign-in, deployment metadata, operational records and safety-control counters. The privacy summary states that personal data is not sold. Customers may delete deployed apps at any time, and account data access or deletion is handled informally on request by email rather than through a documented procedure.
Reuse rights
The privacy summary limits use of collected data to operating, securing, debugging and improving the hosted service and to administering subscriptions, pooled usage limits and safety controls. On the marketing site itself, Statsig collects page views, clicks, device and browser context, performance data, a pseudonymous per-device identifier and session replay; form fields are masked and the lead form is excluded from replay. Customers keep the right to reuse their own data freely, since their product remains authoritative and Noodle Seed never takes ownership of it. Nothing on the site addresses whether customer data is used to train AI models, in either direction, and the platform states that tool-call payloads and the content a customer server processes are not retained as logs.
Data retention & training
Hosting summary
The hosted service runs on Google Cloud, described in the privacy summary as processing data on the vendor's behalf for compute, storage and key management. Statsig processes website analytics and session replay data for the marketing site. Beyond those two names, the site declares neither a hosting country nor a region, and publishes no jurisdiction statement, no transfer mechanism and no data residency option in its general documentation. The domain resolves to a United States address on a Google anycast node, which describes the content delivery edge rather than where customer data actually lives. Packaged assets are served from a separate asset domain, kept apart from MCP and OAuth traffic. For pilots, the security page requires choosing a connectivity model among managed SaaS, a customer-dedicated cloud project, a private cluster, a customer-hosted runner, a secure relay or a hybrid arrangement, and states that internal systems should fail closed if the chosen path is unavailable.
Things to keep in mind
Risks and trade-offs to weigh before adopting Noodle Seed.
- Compliance is claimed through a footer badge and structured data, not through documentation: a GDPR badge and a SOC 2 Type 1 claim sit above a privacy policy with no GDPR section at all
- The public beta status is real and the terms disclaim all warranties, so a workflow your customers depend on rests on a service whose limits may change before general availability
- No retention period, no hosting country and no data processing agreement are published, which will block most procurement and security reviews
- The subprocessor list names only Google Cloud and Statsig and the site itself calls it incomplete, so the true chain of processors is unknown
- Session replay runs on the marketing site through Statsig; form fields are masked and the lead form excluded, but replay is active nonetheless
- Handing agent access to a live product means agent traffic can trigger real actions, so the per-tool scopes, roles and policy limits deserve deliberate configuration rather than defaults
- A single email address covers legal, privacy and support, with no postal address or phone number, which leaves little recourse if something goes wrong
Setup & Integrations
Technical difficulty
Two very different levels. The developer route expects real engineering: TypeScript, Zod schemas, an understanding of OAuth or OIDC, scopes and roles, and an existing API or MCP surface to connect. Coding-agent plugins for Claude Code, Codex and Cursor shorten the authoring work, and everything can be validated and tested locally without an account. The no-code route asks only for a website address, with automatic extraction and a visual editor, though publishing to the ChatGPT app store waits on platform review. Fixed-scope implementation packages exist for teams that would rather have the first workflow launched for them.
Deployment
Integrations
Supported languages
Behind Noodle Seed
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What does Noodle Seed actually do?
Do I need to be a developer to use it?
Why not simply build an MCP server myself?
What does it cost?
Is there a permanent free plan?
Which AI hosts are supported?
How mature is the product?
Where is my data hosted?
Is customer data used to train AI models?
Is there an affiliate or partner programme?
Should you pick Noodle Seed?
Noodle Seed is a genuinely technical product with an unusual amount of evidence behind it: a complete TypeScript definition model, a three-command CLI, published code samples, a compilation chain described end to end, a security page that names ten distinct control areas, and around forty customer applications that can be opened and used in the ChatGPT app directory. The candour about maturity is worth noting, because it is rare: the site labels itself a public beta, the documentation marks each capability shipped, partial or roadmap, and the security page states plainly which controls sit in private preview rather than blurring the line.
The gap is on the legal and compliance side, and it is wide. A GDPR badge appears in every footer and the structured data claims SOC 2 Type 1 certification, yet the privacy policy carries no GDPR section, no retention period, no hosting jurisdiction and no data processing agreement, and describes itself as a provisional beta summary. There is no postal address, no phone number, and one email address covers legal, privacy and support. For a product whose entire pitch is governance, that asymmetry is the thing to raise first in an evaluation.
The natural buyer is a SaaS vendor with one customer workflow already worth money and a real request to expose it to an agent. For that buyer, the free plan and the no-account local development make the cost of trying close to zero, which is the honest way to evaluate it. Anyone with procurement, security review or regulatory obligations should get the compliance documentation in writing before committing, and should expect the two product lines and their two separate price ladders to need clarifying with the vendor.
- Choosing a selection results in a full page refresh.
- Opens in a new window.