openserverless-mcp
OfficialAdds Milvus vector database access to an OpenServerless endpoint by injecting the Milvus client into the endpoint's context.
Adds MongoDB support to an OpenServerless endpoint by injecting MongoDB client and database objects into the endpoint's context.
Adds PostgreSQL support to an OpenServerless endpoint by injecting a PostgreSQL client into the endpoint's context.
Adds Redis support to OpenServerless endpoints by injecting Redis client and prefix variables, and uses Redis-backed opaque sessions for authentication.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@openserverless-mcp Create a new endpoint named v1/hello-world"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
openserverless mcp
An MCP server for OpenServerless. It exposes the OpenServerless action tools
(previously the @opencode-ai/plugin tools under trustable-app/tools/) over the
Model Context Protocol so any MCP-capable agent can drive them.
Tools
Tool | Description |
| Idempotently create an API endpoint ( |
| Run |
| Add a library to an endpoint's |
| Wire a |
| Check secret presence and endpoint bindings without reading its value. |
| Atomically bind the same secret to multiple endpoints. |
| Atomically remove an obsolete generated binding without reading or deleting the secret. |
| Atomically wire Redis into all token-issuing, protected/session, and logout endpoints; never writes |
| Add bucket-scoped S3 to an endpoint's context ( |
| Add PostgreSQL ( |
| Add Redis ( |
| Add Milvus vector DB ( |
| Add MongoDB ( |
The endpoint argument is either name (uses the v1 package) or
package/name. Both segments must start with a letter and contain only letters,
numbers, and hyphens. Use flat hyphenated names such as
v1/employees-photo; underscores, spaces, nested routes, and forms such as
v1/employees_photo or v1/employees/photo are invalid.
Repeating action_new for a compatible existing endpoint is a successful check
that leaves its files unchanged; incomplete paths and visibility conflicts are
reported as MCP errors.
Secret values are never returned by these tools. Missing secrets and invalid
endpoints are MCP errors (isError: true), so clients cannot mistake an
incomplete binding for a successful tool call. action_add_secret remains as
the single-endpoint compatibility tool; use secret_bind when several actions
must share one user-configured credential.
Application .env and .env.production files are owned by Trustable's
user-facing configuration flow. The MCP performs only value-free presence
checks needed for binding validation and never creates, edits, imports,
synchronizes, regenerates, or automatically populates those files. A missing
variable must be added by the user through Trustable.
Application authentication uses Redis-backed opaque sessions. Create login,
registration, me/session, every protected endpoint, and logout, then call
auth_setup once with those complete endpoint sets. Store token-to-identity
mappings under keys derived from ctx.REDIS_PREFIX with a bounded TTL; never
use JWT or an application signing secret as a substitute.
Related MCP server: OpenMesha
Working directory
All path-based tools operate on paths relative to the process working
directory: they read/write packages/<pkg>/<name>/... and .env. The server
must therefore be launched with the user's app checkout as its CWD (this is how
opencode launches type: "local" MCP servers).
Layout
src/
index.ts entrypoint — registers every tool over stdio
lib.ts shared helpers (endpoint parsing, connector injection, types)
secrets.ts secret name, .env, status, and atomic binding helpers
tools/
new.ts action_new
invoke.ts action_invoke
requirements.ts action_requirements
add-secret.ts action_add_secret
add-s3.ts action_add_s3
add-postgresql.ts action_add_postgresql
add-redis.ts action_add_redis
add-milvus.ts action_add_milvus
add-mongodb.ts action_add_mongodb
secret-status.ts secret_status
secret-bind.ts secret_bind
secret-unbind.ts secret_unbind
auth-setup.ts auth_setupEach file under tools/ default-exports a Tool (defineTool({ name, config, handler }));
index.ts imports them all and registers them in a loop. To add a tool, drop a
file in tools/ and add it to the array in index.ts.
Run
Requires Node 18+ and ops on the PATH. TypeScript runs directly via tsx.
npm install
npm start # tsx src/index.ts — speaks MCP over stdioWire into opencode
Add to the app's opencode.json:
{
"mcp": {
"openserverless": {
"type": "local",
"command": ["npx", "tsx", "/path/to/mcp/src/index.ts"],
"enabled": true
}
}
}Available Tools
13 toolsaction_add_milvusA
Add Milvus vector DB connection to an endpoint's context. Provides ctx.MILVUS.
Inputs: endpoint (required).
action_add_mongodbB
Add MongoDB connection to an endpoint's context. Provides ctx.MONGODB_CLIENT and ctx.MONGODB.
Inputs: endpoint (required).
action_add_postgresqlA
Add PostgreSQL connection to an endpoint's context. Provides ctx.POSTGRESQL.
Inputs: endpoint (required).
action_add_redisA
Add Redis to one endpoint as ctx.REDIS and ctx.REDIS_PREFIX. For authenticated pages, every login/registration, me/session, protected-resource, and logout endpoint must be wired; use auth_setup to configure the complete set atomically. Authentication uses opaque random tokens stored server-side in prefix-safe Redis keys with a bounded TTL, never JWT or an application signing secret.
Inputs: endpoint (required).
action_add_s3A
Add a bucket-scoped S3 connection to an endpoint's context. Provides ctx.S3_CLIENT, ctx.S3_DATA, ctx.S3_WEB, ctx.S3_PUBLIC. Never call list_buckets; verify read/write with put_object, get_object plus content comparison, and delete_object in ctx.S3_DATA.
Inputs: endpoint (required).
action_add_secretB
Add a secret to an endpoint's context. The secret must exist in .env.
Inputs: secret (required), endpoint (required).
action_invokeA
Invoke an API action and return its JSON result. Executes ops invoke with the given endpoint and key=value parameters.
Inputs: endpoint (required), params.
action_newA
Create a new API endpoint. Creates the action folder with main.py and module file.
Inputs: endpoint (required), public.
action_requirementsA
Add a library to an endpoint's requirements.txt. Skips if the library is preinstalled.
Inputs: library (required), endpoint (required).
auth_setupA
Atomically add Redis session wiring to the complete authentication surface: token issuers (login/registration), token validators (me/session and every protected resource), and logout. This tool never reads or writes .env. Generated auth must use bcrypt password verification plus a cryptographically random opaque token stored as a prefix-safe ctx.REDIS_PREFIX key with a bounded TTL; protected endpoints derive identity from Redis and logout deletes the record. JWT and application signing secrets are forbidden.
Inputs: token_endpoints (required), logout_endpoints (required), protected_endpoints (required).
secret_bindA
Atomically bind one existing .env secret to multiple OpenServerless endpoints as ctx..
Inputs: secret (required), endpoints (required).
secret_statusA
Report whether a secret exists and is bound to selected endpoints without reading or returning its value.
Inputs: secret (required), endpoints.
secret_unbindA
Atomically remove one generated secret binding from multiple OpenServerless endpoints without deleting the secret value.
Inputs: secret (required), endpoints (required).
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 13 tools
Most tools are clearly separated by integration target or resource: Milvus, S3, Postgres, Redis, MongoDB, secret status/bind/unbind, invoke, and auth setup. The main overlap is between action_add_secret and secret_bind, which both attach existing .env secrets to endpoint contexts; their descriptions clarify single vs. multi-endpoint scope, but the boundary could still cause misselection.
The action_ and secret_ prefixes create a readable grouping, and action_add_* is used consistently for integration setup. However, action_new and action_requirements break the verb_noun pattern, and action_add_secret sits awkwardly alongside the secret_* family, making the naming somewhat mixed though still understandable.
With 13 tools, the server is well scoped for an OpenServerless configuration platform. Each tool covers a distinct setup, invocation, secret-management, or auth task, and there are no obvious filler or redundant tools.
The set covers creating and invoking endpoints, adding dependencies and data connections, managing secret bindings, and setting up auth. However, there are notable gaps: no way to list, update, or delete endpoints, and no way to remove an added integration or connection, which can leave agents in dead ends.
Maintenance
Related MCP Connectors
Build and run AI workflows, agents, data tables and 1000+ API integrations from any MCP client.
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
Your org's AI agents, tasks, runs, search, and brain files as MCP tools and resources.
MCP-first control plane for ProAgentStore agents and private instances.
Related MCP Servers
- FlicenseNot gradedqualityCmaintenanceEnables coding agents to create, deploy, invoke, and monitor serverless functions using 12 MCP tools, with support for multiple runtimes and remote MCP access.58 npm-
- AlicenseAqualityBmaintenanceEnables MCP clients to manage and orchestrate autonomous agents, budgets, and multi-agent workflows within the OpenMesha agentic operating system.91Apache 2.0

meshive-mcpofficial
AlicenseAqualityBmaintenanceEnables MCP-capable agents to manage Meshive GPU Cloud resources, including account, workspaces, pods, storage, GPUs, templates, serverless deployments, tasks, assets, machines, and billing history.44Apache 2.0- AlicenseBqualityCmaintenanceEnables MCP-compatible clients to manage AI agent teams, including agents, Kanban issues, projects, goals, approvals, budgets, and raw API access.30MIT