For AI agents · MCP
MCP server
This site hosts a Model Context Protocol server at /mcp, so an AI agent writing Bsqlite code
reads the library itself - its compiled API documentation, its guides, the capabilities its storage
providers declare, its own storage selection and SQL guards, and these demo pages - instead of guessing at
an API it half remembers. Every tool is read-only, and none of them executes SQL.
Point your agent at it
The endpoint is plain Streamable HTTP, so any MCP client connects with a URL and nothing else:
this site's own https://bsqlite.bitplatform.dev/mcp, the deployed https://bsqlite.bitplatform.dev/mcp,
or https://bitplatform.dev/mcp, which serves the tools of every bit library from one
address. The same tools are plain GET endpoints under /api/mcp/..., which is
what the panels below call.
{
"mcpServers": {
"bsqlite": {
"type": "http",
"url": "https://bsqlite.bitplatform.dev/mcp"
}
}
}claude mcp add --transport http bsqlite https://bsqlite.bitplatform.dev/mcp
Where will the database open?
PlanBsqliteStorageSelection runs the library's own resolver, migrator and session
factory - composed exactly as AddBitBsqliteServices composes them - against stand-in
providers that answer their probe the way you describe the browser. The first-party backends keep
the capabilities their provider types declare. Nothing is opened.
Will this statement run on every backend?
CheckBsqliteSql puts SQL to the library's own static rules: the guards each backend's
declared capabilities switch on, and the table scan a live query is keyed by. It is pattern
matching over the text - nothing is executed, on the server or anywhere else.
Everything at once, with the next call to make
An agent rarely knows which corpus holds the answer. SearchBsqlite covers every
document, every public member of both packages, the storage backends, the setup guides, the labs
and the demo's sources at once - and every hit carries the exact follow-up call that returns it in full.
Tools
The list loads once the page is interactive.
Prompts and resources
Prompts are the workflows - the order to call the tools in for a whole job - because the failure mode of an
agent holding a shelf of tools is calling them in an order that skips the check which would have caught
the bug. Resources are the same knowledge as bsqlite:// URIs, for clients that attach
documentation up front; they read the same catalogs the tools do.
The same tools over plain HTTP
Open one in a browser tab to read exactly what an agent is handed.
| Tool | Example |
|---|---|
SearchBsqlite |
api/mcp/SearchBsqlite?query=live query |
GetBsqliteSetupGuide |
api/mcp/GetBsqliteSetupGuide?scenario=wasm |
GetBsqliteDocument |
api/mcp/GetBsqliteDocument?name=storage§ion=Selection |
GetBsqliteApi |
api/mcp/GetBsqliteApi?typeName=BsqliteDatabase |
GetBsqliteStorageProviders |
api/mcp/GetBsqliteStorageProviders?provider=opfs |
PlanBsqliteStorageSelection |
api/mcp/PlanBsqliteStorageSelection?unavailable=opfs |
CheckBsqliteSql |
api/mcp/CheckBsqliteSql?sql=PRAGMA page_size = 8192 |
GetBsqliteSourceFile |
api/mcp/GetBsqliteSourceFile?path=Demo/Client/Components/SessionLifecycle.razor |