QuantConnect MCP Server

Official

The official QuantConnect MCP server that lets AI operate the quantitative trading platform directly.

How to install ↓ Compare GitHub source repository ↗

Category: Dev Tools #362 of 438
Stars: ★ 79 Popular
Transport: stdio (local process)
Runtime: Docker
Credentials: API key / credential required
License: Apache-2.0
Last commit: 5 months ago
Tools: 64

39FMRS · D

The official QuantConnect MCP server offers 64 tools covering the full quant workflow: project management, compilation, backtesting, optimization, live trading, and Object Store. Dockerized for cross-platform use and officially maintained, it is powerful but warrants caution with live-trading permissions; note the repo is currently deprecated in favor of the VSCode-embedded MCP.

Strongest · Setup experience 13/20 Weakest · Security and permissions 4/20

Reliability

6/20

Security and permissions

4/20

Maintenance

5/20

Documentation

11/20

Setup experience

13/20

Why each score

Reliability 6/20: The README documents 64 tools with parameter tables and short descriptions in a consistent format, but the repository provides no verifiable CI workflow or committed tests to confirm the documented behavior matches the implementation (per static calibration, this caps the score at 12). More importantly, several tools that are clearly read-only by name and description (read_live_algorithm, read_live_chart, read_live_logs, read_live_portfolio, read_live_orders, read_live_insights, list_live_algorithms) are annotated as 'modifies its environment' and 'may perform destructive updates' — a direct contradiction that suggests the manifest annotations are auto-generated and unverified, undermining confidence that declared tool behavior is accurate. No error-handling or dependency-control details are given.
Security and permissions 4/20: This server bridges directly to a real-money live trading account using the user's own API token, exposing liquidate_live_algorithm (irreversible liquidation of live positions), create_live_algorithm (deploys live trading), broadcast_live_command (broadcasts a command to every live algorithm in an organization), and multiple irreversible delete_* tools (projects, files, backtests, optimizations, object store). Nowhere in the documentation is there any mention of confirmation, dry-run, or scoping safeguards for these high-stakes/irreversible operations, and broadcast_live_command has an especially wide blast radius (entire organization). This matches the rubric's red-line condition for 'irreversible delete/pay/deploy/remote-command defaults without confirmation,' so the security score is placed in the red-line range. Credentials are passed via environment variables and no real secrets appear in the examples, which is the one mitigating factor.
Maintenance 5/20: The README's top banner explicitly states 'This repository is deprecated for now' and redirects users to a different, VSCode-embedded implementation — signaling that the maintainers have effectively stopped treating this repo as the recommended path forward, even though it is not formally archived on GitHub. 77 stars and 15 open issues indicate past interest, but there is no evidence of active issue triage or dependency updates going forward. Apache-2.0 licensing gives clear ownership, which is a positive, but the deprecation notice substantially undercuts confidence in ongoing maintenance.
Documentation 11/20: The README is reasonably layered: install steps, a Claude Desktop JSON config example, parameter tables for all 64 tools, and a debugging section (build/logs/inspector) are all present, making onboarding straightforward. However, it omits cost disclosure (backtests, optimizations, and live deployments on QuantConnect typically incur real costs), rate limits, the scope of permissions granted by the API token, and detailed troubleshooting for failure scenarios; it also doesn't clarify how much of the documentation still applies given the deprecation notice. The tool-annotation contradictions noted above further erode documentation reliability.
Setup experience 13/20: The setup path is clear: install Docker Desktop and Claude Desktop, edit the config file, supply user ID and API token, restart the client — few steps, a concrete JSON example, and explicit linux/amd64 vs linux/arm64 platform guidance for cross-platform use. Because no verifiable execution evidence (e.g., a passing CI build) is provided, static calibration caps this dimension at 15; the deprecation notice also introduces uncertainty about how long this setup path will remain valid.

Static review · not runListed 2026-09-07

Read the FMRS scoring method →

Fit and risk

What it can accessUses the networkChanges third-party account data

Best for

QuantConnect platform users who want MCP clients like Claude Desktop to operate their cloud quant workflow via natural language
Quant traders seeking automated backtesting, optimization, and live deployment management

Not for

Traders who do not use the QuantConnect cloud platform
Users seeking a local non-Docker setup who don't want to configure API credentials (the repo is deprecated; the VSCode-embedded MCP is preferred)

Required permissions

Requires QuantConnect user Id and API token (provided via environment variables)
Can create, modify, and delete cloud projects, files, backtests, and optimizations
Can deploy, stop, and liquidate live algorithms and broadcast commands (real-money risk)
Can authorize external brokerage and data provider connections

Risks and side effects

Live trading tools (e.g., liquidate_live_algorithm, broadcast_live_command) can cause real financial losses
A leaked API token exposes your account and strategy code
Delete tools are destructive and may be unrecoverable
Repeated calls to some create tools produce duplicate resources
The repository is deprecated and may no longer be maintained

Setup

Before you start

Runtime：Docker

QUANTCONNECT_USER_ID required Your QuantConnect user ID, obtained from your profile per the 'Request API Token' docs page.

QUANTCONNECT_API_TOKEN requiredsecret Your QuantConnect API token (secret), requested on the same 'Request API Token' page.

Other optional settings (1)

AGENT_NAME optional Optional agent name used to track the request source when running multiple agents; defaults to 'MCP Server'.

Install and open Docker Desktop
Install and open Claude Desktop
In Claude Desktop, go to File > Settings > Developer > Edit Config
Edit claude_desktop_config to add a 'quantconnect' server entry running the quantconnect/mcp-server Docker image with QUANTCONNECT_USER_ID, QUANTCONNECT_API_TOKEN, and AGENT_NAME environment variables, choosing --platform linux/amd64 or linux/arm64 for your chip
Restart Claude Desktop, which automatically pulls the image from Docker Hub and connects. Update with 'docker pull quantconnect/mcp-server'. Set a unique AGENT_NAME per agent when running multiple agents.

claude_desktop_config.json

{
  "mcpServers": {
    "quantconnect": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "QUANTCONNECT_USER_ID",
        "-e",
        "QUANTCONNECT_API_TOKEN",
        "-e",
        "AGENT_NAME",
        "--platform",
        "<your_platform>",
        "quantconnect/mcp-server"
      ],
      "env": {
        "QUANTCONNECT_USER_ID": "<your_user_id>",
        "QUANTCONNECT_API_TOKEN": "<your_api_token>",
        "AGENT_NAME": "MCP Server"
      }
    }
  }
}

Shown for Claude Desktop. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json

{
  "servers": {
    "quantconnect": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "QUANTCONNECT_USER_ID",
        "-e",
        "QUANTCONNECT_API_TOKEN",
        "-e",
        "AGENT_NAME",
        "--platform",
        "<your_platform>",
        "quantconnect/mcp-server"
      ],
      "env": {
        "QUANTCONNECT_USER_ID": "<your_user_id>",
        "QUANTCONNECT_API_TOKEN": "<your_api_token>",
        "AGENT_NAME": "MCP Server"
      }
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal

claude mcp add quantconnect -e 'QUANTCONNECT_USER_ID=<your_user_id>' -e 'QUANTCONNECT_API_TOKEN=<your_api_token>' -e 'AGENT_NAME=MCP Server' -- docker run -i --rm -e QUANTCONNECT_USER_ID -e QUANTCONNECT_API_TOKEN -e AGENT_NAME --platform '<your_platform>' quantconnect/mcp-server

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

After restarting Claude Desktop, the tool list should include tools like read_mcp_server_version and list_projects; calling read_mcp_server_version and getting the running server version back proves the connection works.

Troubleshooting

Verify Docker Desktop is running and QUANTCONNECT_USER_ID/QUANTCONNECT_API_TOKEN are configured
On ARM chips (e.g., Apple M-series), add --platform linux/arm64
Restart Claude Desktop to trigger image pull and connection
Run read_mcp_server_version to confirm the server version and update via docker pull if needed
For debugging, log via sys.stderr to mcp-server-quantconnect.log, or inspect with npx @modelcontextprotocol/inspector

Things to try

Once connected, you can ask your AI assistant things like:

List all projects in my QuantConnect account.
Create a Python project called AlphaResearch in my default organization and add a main.py file to it.
Create a compile job for my most recent project, then run a backtest and show me the results summary.
Show the status and portfolio of all my currently running live algorithms.

Tools 64

read_account read-only

Read the organization account status.

create_project writes

Create a new project in your default organization.

read_project read-only

List the details of a project or a set of recent projects.

list_projects read-only

List the details of all projects.

update_project writes

Update a project's name or description.

delete_project destructive

Delete a project.

create_project_collaborator writes

Add a collaborator to a project.

read_project_collaborators read-only

List all collaborators on a project.

Show 56 more tools

update_project_collaborator writes

Update collaborator information in a project.

delete_project_collaborator destructive

Remove a collaborator from a project.

lock_project_with_collaborators writes

Lock a project so you can edit it.

read_project_nodes read-only

Read the available and selected nodes of a project.

update_project_nodes writes

Update the active state of the given nodes to true.

create_compile writes

Asynchronously create a compile job request for a project.

read_compile read-only

Read a compile packet job result.

create_file writes

Add a file to a given project.

read_file read-only

Read a file from a project, or all files if no file name is provided.

update_file_name destructive

Update the name of a file.

update_file_contents destructive

Update the contents of a file.

patch_file destructive

Apply a patch (unified diff) to a file in a project.

delete_file destructive

Delete a file in a project.

create_backtest writes

Create a new backtest request and get the backtest Id.

read_backtest read-only

Read the results of a backtest.

list_backtests read-only

List all the backtests for the project.

read_backtest_chart read-only

Read a chart from a backtest.

read_backtest_orders read-only

Read the orders of a backtest.

read_backtest_insights read-only

Read the insights of a backtest.

update_backtest writes

Update the name or note of a backtest.

delete_backtest destructive

Delete a backtest from a project.

estimate_optimization_time read-only

Estimate the execution time of an optimization with the specified parameters.

create_optimization writes

Create an optimization with the specified parameters.

read_optimization read-only

Read an optimization.

list_optimizations read-only

List all the optimizations for a project.

update_optimization writes

Update the name of an optimization.

abort_optimization destructive

Abort an optimization.

delete_optimization destructive

Delete an optimization.

authorize_connection writes

Authorize an external connection with a live brokerage or data provider.

create_live_algorithm writes

Create a live algorithm.

read_live_algorithm read-only

Read details of a live algorithm.

list_live_algorithms read-only

List all past and current live trading deployments.

read_live_chart read-only

Read a chart from a live algorithm.

read_live_logs read-only

Get the logs of a live algorithm.

read_live_portfolio read-only

Read the portfolio state of a live algorithm.

read_live_orders read-only

Read the orders of a live algorithm.

read_live_insights read-only

Read the insights of a live algorithm.

stop_live_algorithm destructive

Stop a live algorithm.

liquidate_live_algorithm destructive

Liquidate and stop a live algorithm.

create_live_command writes

Send a command to a live trading algorithm.

broadcast_live_command writes

Broadcast a live command to all live algorithms in an organization.

upload_object writes

Upload files to the Object Store.

read_object_properties read-only

Get Object Store properties of a specific organization and key.

read_object_store_file_job_id writes

Create a job to download files from the Object Store and read the job Id.

read_object_store_file_download_url read-only

Get the URL for downloading files from the Object Store.

list_object_store_files read-only

List the Object Store files under a specific directory in an organization.

delete_object destructive

Delete the Object Store file of a specific organization and key.

read_lean_versions read-only

Return a list of LEAN versions with basic information.

check_initialization_errors read-only

Run a backtest for a few seconds to initialize the algorithm and get initialization errors.

complete_code read-only

Show code completion for a specific text input.

enhance_error_message read-only

Show additional context and suggestions for error messages.

update_code_to_pep8 read-only

Update Python code to follow PEP8 style.

check_syntax read-only

Check the syntax of code files.

search_quantconnect read-only

Search for content in QuantConnect.

read_mcp_server_version read-only

Return the version of the running QC MCP Server.

read_latest_mcp_server_version read-only

Return the latest released QC MCP Server version.

Use cases

Let an AI agent create and manage quantitative research projects on QuantConnect's cloud

Write and patch strategy code via AI, then trigger compiles and backtests

Run parameter optimizations and read backtest charts, orders, and insights

Deploy, monitor, stop, or liquidate live trading algorithms

Manage Object Store files and LEAN version information

Supported clients

Claude Desktop

Listed from the project's documentation, not tested by this site.

Overview

The QuantConnect MCP Server is the official Python MCP server maintained by the QuantConnect team, acting as a bridge for AIs (such as Claude and OpenAI o3 Pro) to interact with the QuantConnect cloud platform. Through the QuantConnect API, the AI can update projects, write strategies, compile, backtest, run parameter optimizations, and deploy strategies to production live trading on your behalf. The implementation is tested and dockerized for easy cross-platform deployment, supporting linux/amd64 and linux/arm64. Note: this repository is now deprecated; the preferred way is the MCP embedded in VSCode.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →

NPM Sentinel MCP Server 79 · B

AI-powered NPM package analysis MCP server

★ 18 · Tools 19 Compare with this →

MartinLoop MCP Server 79 · B

An independent control layer that gives AI coding agents budgets, brakes, and receipts

★ 413 Compare with this →

GitHub MCP Server 76 · B

GitHub's official MCP server for managing repos, issues, PRs, and workflows via natural language

★ 33.7k · Tools 15 Compare with this →

Compare with other MCP servers

Source revision 1f0fa4e4f465 Data synced 2026-10-19 Read the FMRS scoring method