MCP Server - MCP Toolkit

The hosted MCP Server gives MCP clients a single Streamable HTTP endpoint for working with MCP Assistant.

Endpoint

Use this URL in MCP clients that support Streamable HTTP:

https://api.mcp-assistant.in/mcp

Connect and manage remote MCP servers from mcp-assistant.in. The hosted endpoint uses those connected servers when MCP clients call MCP Assistant.

What is the hosted MCP Server?

The MCP Server connects AI clients to MCP tools and remote app integrations through the Model Context Protocol. Use it when you want an MCP client to access MCP Assistant without hosting your own server. After connecting remote MCP servers in mcp-assistant.in, the MCP Assistant server can provide access to 100+ MCP servers, including GitHub, Notion, Zapier, Supabase, and other supported services. It also exposes meta-tools for dynamic MCP discovery and a CodeMode tool that executes programs inside a secure sandbox for programmatic tool calling and result processing. This helps avoid expensive LLM tool-calling loops when a task is better handled as a small program.

Who it is for

For MCP Client Users

Connect remote MCP servers in mcp-assistant.in, then use them from MCP clients such as VS Code and Antigravity through one hosted endpoint.

For Developers

Build agents with tool discovery, schema inspection, and CodeMode scripts across connected MCP servers.

What it provides

A hosted MCP endpoint for clients that support Streamable HTTP.
Access to 100+ MCP servers, including GitHub, Notion, Zapier, Supabase, and other supported services.
Meta-tools for dynamic MCP discovery across 100+ supported MCP servers.
Server-side tool execution through MCP Assistant.
CodeMode support for running small programs that call MCP tools inside a secure sandbox.
A simple core flow: list servers, search tools, inspect schema, then run with CodeMode.
Separate workflow tools for saving scripts and inspecting workflow state.

Popular connected servers

Server	Example tasks
GitHub	Create issues, inspect pull requests, search repositories, summarize project activity.
Notion	Search pages, update databases, organize notes, generate workspace summaries.
Zapier	Reach connected app actions and automate cross-app tasks.
Supabase	Query project data, inspect records, and support database-backed workflows.
Slack or Discord	Send messages, post updates, and route workflow notifications.

Example use cases

Personal productivity

Search connected tools and save useful results to Notion.
Summarize recent GitHub activity and send a status update.
Run a small CodeMode script for daily app checks.

Development and operations

Create a GitHub issue from a bug report.
Query Supabase and summarize the result before returning it to the user.
Chain multiple MCP tool calls inside codemode_run to avoid repeated LLM round trips.

Code-based automation

Search for the right connected MCP tool.
Inspect its schema.
Use codemode_run to call one or more tools in a small script.
Return filtered, summarized, or transformed results to your MCP client.

Main tools

The primary hosted MCP flow uses four tools: list, search, inspect, and run. Start here for most MCP clients.

Tool	Category	Description
`list_mcp_servers`	Search	List all connected MCP servers and the number of tools each provides. Use this when `search_mcp_tools` returns no response or irrelevant results.
`search_mcp_tools`	Search	Search connected MCP tools for the authenticated user and return normalized discovery results. Use this first before inspecting schemas or executing tools.
`get_mcp_tool_schema`	Read	Inspect the exact input schema, output schema when available, and CodeMode extraction hint for one connected MCP tool.
`codemode_run`	General	Execute a JavaScript/TypeScript-like script immediately in the CodeMode sandbox with connected MCP tool servers. Use it to validate tool calls, chain multiple calls, batch process results, or transform large outputs without creating a saved workflow.

Workflow tools (Experimental)

[!WARNING] Deprecation Notice: Workflow and scheduling tools are experimental and may be removed at any time without notice.

Workflow and scheduling tools are separate from the main hosted MCP flow. Use them when you want to save scripts or inspect workflow metadata. Queued workflow execution is disabled on the hosted endpoint unless a dedicated worker service is running.

Tool	Category	Description
`workflow_list`	General	List script workflows with schedule and recent run summaries.
`workflow_get`	Read	Get one workflow, including schemas, script code/runtime, and schedules.
`workflow_upsert_script`	General	Create or update a script workflow that runs inside the `@mcp-ts/codemode` sandbox and can call connected MCP tools.
`schedule_upsert`	General	Create or update a UTC 5-field cron schedule. Set `is_enabled` to `false` to save a schedule without enqueueing runs.
`execution_log_list`	General	List recent workflow runs, optionally filtered by `workflow_id`.
`execution_log_get`	Read	Get details for a single workflow run by `execution_log_id`.
`workflow_run`	Optional	Run a saved workflow by enqueueing a BullMQ job. This requires a dedicated worker service and may not be available on the hosted endpoint.
`agent_run`	General	Call an LLM helper that returns structured `summary` and `analysis` output for explanation, interpretation, or recommendation-style reasoning.

Recommended MCP tool flow

When using connected MCP tools through the MCP Server:

Search for candidate tools with search_mcp_tools.
Inspect the selected tool with get_mcp_tool_schema.
Execute the call with codemode_run.

Inside CodeMode, scripts run as the body of an async function. Use return for the final value, and use console.log for debugging. Connected MCP tools are available through namespaced helpers and through callTool(serverId, toolName, args). These helpers return normalized tool results by default; use raw helpers only when you need the original MCP envelope.

Availability and health

The hosted endpoint is designed as a workerless remote MCP service. It keeps the Streamable HTTP MCP server online without running a separate BullMQ worker service. This reduces hosting cost and improves endpoint reliability for MCP clients that primarily need remote tool discovery and codemode_run. Operational health is exposed at:

https://api.mcp-assistant.in/healthz

The service also supports OpenTelemetry traces and metrics through OTLP for production monitoring.

Client configuration

VS Code

{
  "servers": {
    "mcp-assistant": {
      "type": "http",
      "url": "https://api.mcp-assistant.in/mcp"
    }
  }
}

Antigravity

{
  "mcpServers": {
    "mcp-assistant": {
      "serverUrl": "https://api.mcp-assistant.in/mcp"
    }
  }
}

​Endpoint

​What is the hosted MCP Server?

​Who it is for

For MCP Client Users

For Developers

​What it provides

​Popular connected servers

​Example use cases

​Personal productivity

​Development and operations

​Code-based automation

​Main tools

​Workflow tools (Experimental)

​Recommended MCP tool flow

​Availability and health

​Client configuration

​VS Code

​Antigravity

​Related docs