GenHTTP Lambda MCP Server

Start a feature: a place to change a lambda without touching what it has online. It branches off a version - the newest unless base names another - with a copy of its files (its documentation and tests included) and a copy of the lambda's data, and gets an address of its own to try it at. Change it with change_code or write_code and feature (deploy: true puts it online at its preview address), test it there, and merge_feature once it does what was asked. The lambda goes on serving its visitors from the version online the whole time.

Throw a feature away: its files, its preview and its copy of the data. The lambda is not touched. For a feature the user does not want after all.

Rename a feature, change what it says about itself, or move its base. merge_feature refuses a feature that is not based on the newest version, since merging it would undo what was saved after it branched off: bring the newer versions' changes into the feature first, then move base to the newest version here. Nothing checks that the changes really are in - that is up to you.

Create a lambda. Returns its public address and a private editor key, the only way back in. It starts with version 1 - an empty starter, or a copy of a demo. Nothing is online until deploy.

A lambda's status (online version, newest version, expiry, its tier and what it may use there), its open features, the recent versions with what each was asked for and changed, its data (the database, the workspace and the secrets: whether each is on, what it holds, and which secrets the code reads that have no value yet), and one version - or, with feature, that feature: its documentation (docs/: what the app is and why, the technical decisions), how it is tested (tests/), and its files. The documentation comes first and in full up to 20,000 characters; the files come in full while they fit into 30,000 characters - the C# first, then the resources, then the rest of the code - and by name and length after that, with file to read one - a test script or its data too. Read the documentation and the history before changing what you did not write. Also how a demo is read: pass its key from list_demos.

Change how a lambda's editor opens: view 'Simple' shows the app, how it is doing and a box to ask for a change - for an owner who does not write code - and 'Full' every section, the code, files, data, versions and logs included. Only the default: whoever opens the editor can switch for themselves, and that choice stays theirs. Only do this when the user asks for it.

Compile without saving or deploying; returns diagnostics with file and line. Does not build the handler, so deploy can still refuse a route whose return type cannot be served.

Save every file, replacing the previous set: as a new version of the lambda, or - with feature - into that feature. A version is its code and its resources. The code is every file not in resources/: its .cs files are compiled, in any folder, as in a C# project - lambda.cs returns the handler, the others hold types - and every other file is kept with it and never compiled or served: docs/product.md, docs/decisions.md, tests/README.md and the tests' scripts and data are what is written about the program, and whatever else - what a front end is built from, a tool's configuration - is laid out as you like (platform_guide, code). resources/ is what the lambda reads and serves while it runs, as Resources: the whole front end (pages, scripts, styles, icons) goes there, as part of the program - resources/web/index.html is served by Resources.App("web"). What the lambda keeps at runtime is data, never files here: records and accounts in the database (a DbContext of Entity Framework Core on Database.GetConnection(), used synchronously; its schema as Evolve migrations in resources/migrations/), uploads in the workspace - and so is a large input file such as a model or a dataset (upload_file). Send the documentation and tests with the code: written with a new lambda, updated with every change, in proportion to the lambda - a few lines and one quick check for a small one. Say why with specification and change. deploy: true publishes in the same call - a version at the public address, a feature at its preview address. To send only what changes, use change_code. To change a lambda that is already in use, work in a feature.

Remove a secret. Code that reads it with Secret.Read fails from its next call on. Only when the user asks for it, or a secret you stored yourself is no longer used. With feature, from that feature's copy instead.

Store an API key, password or token under a name the code reads with Secret.Read("NAME") - never put one in the code, a resource, the workspace, a log line or a note. The value is sealed and can never be read back, by you or the owner; only the lambda reads it, from its next call on, without a deploy. Only set a value the user gave you for this: when you do not have it, write the code with Secret.Read and tell the user to set it under Data > Secrets in the editor - read_lambda and list_secrets list what is still missing. Secrets must be switched on first (enable_data). With feature, into that feature's copy (a sandbox key to try it with). Up to 100 secrets of 32 KB each.

How people reach a lambda, as its owner sees it in the editor: visitors, pages opened and requests per day, and the pages, referring sites, countries, devices, browsers, operating systems and crawlers (search engines, AIs, link previews) behind them. Counted by day in UTC and written down every few minutes; a year is kept, of which a free lambda shows the last 7 days and a premium one all. Off until switched on - nothing is counted before. enable: true switches it on, only when the user asks; switching off deletes what was counted and is the owner's to do in the editor. Use it to answer what the user asks about their visitors, or to see which pages matter before changing them.

Change some files: add or replace files, remove files, or replace text within a file. Everything not named stays as it is, so there is no need to resend unchanged files. With feature, the change is saved into that feature - the way to work on a lambda that exists: change it as often as you like, deploy: true to try it at the feature's preview address, merge_feature when it is right. Without feature, the newest version is changed and saved as a new version. Update what the change affects in docs/ and tests/ in the same change - an edit of a passage is enough. check: true compiles without publishing. Code that does not compile is saved but never goes online.

The lambda's database - its records, shared by every version: whether it is switched on, how full it is, and its tables and views with their columns and how many rows each holds. With table, a page of that table's rows, newest first, up to 100 at a time. Read only: the lambda writes it, through Database.GetConnection(). Use it to see that migrations were applied and records were written. With feature, that feature's copy.

Demos this platform keeps online, each a finished lambda showing one way to build something: a REST API over records, registration and login, a websocket game, uploads, live updates. Their keys are public and read only: read the closest one with read_lambda (and list_files, read_logs) before writing similar code. create_lambda with a demo's id as template starts from a copy.

Deploy (publish) a saved version so it goes live at its public address - the newest by default. With feature, deploy that feature's files to its own preview address instead, against its copy of the data, leaving the lambda alone. Returns diagnostics on failure, and whatever was online stays online.

Switch on a kind of data the lambda needs and does not have: 'database' (a SQLite database for its records - off until switched on, which makes it, empty), 'secrets' (off until switched on) or 'workspace' (on unless the owner switched it off). Do it when what you build needs it - records need the database, an API key for a service it calls needs secrets. Takes effect at once, without a deploy; switching the database on starts the lambda again on its next request, so its migrations run. There is no tool to switch one off: that deletes what it held, and is the owner's to do in the editor.

Remove a file, or a folder with its contents, from the workspace - the lambda's data, shared by every version. No deploy or rollback brings it back. With feature, from that feature's copy instead.

The lambda's data: whether its workspace is switched on, and every file in it with size and last write - the same whichever version is online. With feature, that feature's copy of it. The code and resources of a version are in read_lambda.

What a deployed lambda has been doing: its recent requests and how they were answered, what it printed, the errors it threw with their stack traces, and how much traffic it has had in the last hour and day. With feature, what that feature's preview has been doing instead, kept apart from the lambda's visitors. Call it after deploying to see that it works, and first when something is reported broken.

Make a feature's files the next version of the lambda, and delete the feature with its preview and its copy of the data - the lambda's own data is not touched. Its docs/ and tests/ become the version's documentation and tests, so bring them up to date in the feature first, and run its tests against the preview. Refused while the feature is not based on the newest version (update_feature says how to get it there), and while its code does not compile. deploy: true puts the new version online at once; otherwise deploy it when the user wants it live.

Publish a lambda's source code at /source/{publicKey} for anybody to read, star and download as a .NET project, change the license it is published under, or take it down. What is published is the program and what is written about it - every version's code, resources, documentation and tests, and the history of what each changed - never its data: not the database, not the workspace, not a secret's value. Not part of building: only do this when the user asks for it, and ask which license if they did not say (MIT unless they want another). With only privateKey it returns whether it is published. Once published, everything in every version is public, older ones included: keep keys, passwords and personal data out of the files - in secrets and the database.

How this platform works - read it first: what a version, a feature and data are and how long each lives, what is code and what is a resource, the documentation and tests every version keeps, what the snippet returns, what is imported, what is refused, limits and terms.

The lambda's secrets by name - never their values: whether secrets are switched on, what is stored and when, which names the code reads with Secret.Read and have no value yet (missing) - what the user still has to set - and which it only uses when they are there, checked with Secret.Exists (optional). With feature, that feature's copy.

List a lambda on the public showcase page, change its entry, or take it off. Not part of building: only do this when the user asks for it. With only privateKey it returns the current entry. An entry needs a title, a description and a picture (a screenshot or short GIF of the lambda in use); it is listed while the lambda is online. Write plainly and concretely: what it is and what a visitor can do with it. No marketing language, no superlatives, no exclamation marks, no emoji.

Write a file to the lambda's workspace - its data, which every version shares and no deploy, rollback or merge touches. With feature, to that feature's copy of the data instead, for trying things out without touching the real one. For files the lambda works with at runtime (pictures people will browse) and large input files that are not program: a model, a dataset, media. Records belong in the database, where the code writes them. Takes effect at once, without a deploy. Not for the front end: pages, scripts and styles are the program and belong in the version's resources (write_code).