MCP Reference All tools and resources exposed by openserve’s Model Context Protocol server. Built for Claude Code, Cursor, and any MCP-compatible client.
14 tools5 resourcesstreamable HTTP · SSE API-key auth · stable
Connection§ /connection Point your MCP-compatible client at the openserve endpoint. Every request is authenticated with an API key generated from your dashboard.
Endpoint
https://api.openserve.dev/mcp streamable http
Auth
Authorization: Bearer YOUR_API_KEY
Region
sfo-1 · hnd-1 · ams-1 (auto-routed)
Transport
HTTP/2 · keep-alive · SSE chunked
~/.claude/mcp.json
json {
"mcpServers" : {
"openserve" : {
"type" : "streamable-http" ,
"url" : "https://api.openserve.dev/mcp" ,
"headers" : {
"Authorization" : "Bearer YOUR_API_KEY_HERE"
}
}
}
}
Once connected, your client will see openserve as an available MCP server. List its capabilities with the standard MCP tools/list and resources/list methods, or just browse below.
Resources§ /resources · 5 Resources are read-only context documents that LLMs read before writing code. The static resources (sdk-docs, rules, examples, shared-file-types) require no arguments. The dynamic project-context resource is scoped to a project and requires auth.
RES scf://sdk-docs SDK documentation: auto-injected globals, helper APIs, available imports, and function contract. text/markdown RES scf://rules SCF code rules: forbidden APIs, forbidden patterns, import constraints, and structural requirements. text/markdown RES scf://examples Five complete example functions: DB read/write, Redis caching, external APIs, validation, errors. text/markdown Five complete example functions showing SCF patterns. Use these as templates when creating new functions — they cover the patterns that account for ~90% of cloud-function code on openserve.
Includes
· Read/write Postgres · Redis caching · External fetch with retry · Parameter validation with zod · Structured error responses Sample call
{
"method" : "resources/read" ,
"params" : {
"uri" : "scf://examples"
}
}
RES scf://shared-file-types Catalog of all shared-file categories supported by SCF, with purpose and examples for each. application/json RES scf://project/{projectId}/context Per-project context: functions, shared files, and metadata. Requires authentication. text/markdown Tools are operations the LLM can call. Grouped by domain: projects, functions, shared files, and database. Every tool returns a JSON object — schemas below.
Projects / 4 tools GET scf_list_projects List all active projects owned by the authenticated user. Returns counts of functions and shared files so you can understand what each project contains. Use this to find the projectId needed by other tools.
Arguments
No arguments — call with an empty object.
Returns
projectId string
Stable identifier — use this in subsequent calls.
name string
Human-readable project name.
description string · nullable
Optional short description set by the owner.
functionCount integer
Number of cloud functions in this project.
sharedFileCount integer
Number of shared files in this project.
Example response
[
{
"projectId" : "prj_8jK2" ,
"name" : "focus-timer" ,
"description" : "Pomodoro w/ streaks & sync" ,
"functionCount" : 7 ,
"sharedFileCount" : 3
} ,
{
"projectId" : "prj_q4M9" ,
"name" : "penned" ,
"description" : "Journal that writes back" ,
"functionCount" : 12 ,
"sharedFileCount" : 5
}
]
POST scf_create_project Create a new project — the top-level container for cloud functions, shared files, and data. Checks your plan quota first. GET scf_get_project_description Read a project's freeform markdown description (goals, architecture, conventions) — metadata kept separate from code. PATCH scf_update_project_description Write or replace a project's markdown description. Pass an empty string to clear it.
Functions / 8 tools POST scf_create_function Create a cloud function from ESM source and auto-deploy it. Runs the scf://rules validator first; the functionId is returned even if deploy fails. GET scf_list_functions List a project's functions with deploy status and last-deployed time. Browse before creating (avoid name clashes) or to find a functionId. GET scf_get_function_code Read a function's current source and metadata (read-only). Call before scf_update_function to preserve working logic. PATCH scf_update_function Save new code for a function, snapshot a version, and redeploy. The update always persists even if deployment is skipped. POST scf_deploy_function Run the full deploy pipeline (security check → compile → write → notify) on a function's saved code without changing it. POST scf_invoke_function Invoke a deployed function with a body forwarded as reqBody. 12s timeout — use for testing during agent coding loops. GET scf_get_function_logs Retrieve a function's deploy history — status, error, and timing per attempt, newest first. Diagnose failed deploys. POST scf_ai_generate_function Generate a cloud function from a plain-English description and auto-deploy it. SDK-aware; requires the AI service to be configured.
Shared files / 1 tool POST scf_create_shared_file Create a shared module (model / util / constant / service / middleware) importable by any function via require() injection at deploy time.
Database / 1 tool POST scf_setup_collection Register a namespaced MongoDB collection ({projectId}-{name}) and get the exact dbHelper call to use in your function code. Errors§ /errors Errors follow the standard MCP error envelope. Every error has a numeric code, a stable tag for programmatic handling, and a human message.
{
"error" : {
"code" : -32602 ,
"tag" : "forbidden_import" ,
"message" : "Function imports 'fs' — SCF rules forbid filesystem access." ,
"hint" : "Read scf://rules for the full list of forbidden APIs."
}
}