AI agents for investment research
Research assets and build, test, and maintain allocation models with a compatible AI client. Inspect saved code, inputs, runs, and current allocations.
Try this workflow
Give your agent a persistent research workspace
Example starting prompt
Use ENSEMBLE to build and backtest a monthly 60/40 allocation. Inspect the code and data assumptions, then return the model link and current target weights.
MCP endpoint: https://api.ensemble.markets/mcp
- 01
Connect
Create an account and connect an MCP client with OAuth or a dashboard API key.
- 02
Research
Ask your agent to build and test a specific allocation model, then inspect the result.
- 03
Continue
Open the same saved model in the web app, API, MCP, or CLI and compare later runs.
No card required. Starting credit expires after 7 days. Builds and runs use metered credit; continued usage requires available credit or a funded plan. Backtests are hypothetical. ENSEMBLE produces research and allocations; it does not place trades.
An AI assistant can help express an investment idea, but the useful output is the saved model and its supporting evidence. ENSEMBLE lets people and their agents research assets, build and test allocation models, inspect their work, and retrieve allocations later. An allocation describes how investments are divided among assets or other models.
Start with a workflow
Use the finance MCP guide to understand the available tools, data coverage, and authentication. The Claude walkthrough shows example requests from an asset check through an isolated parameter test.
For a browser-first approach, follow how to backtest a portfolio. The web app, API, CLI, and connected agents work with the same models; an agent is optional.
What stays with the model
Models retain source, declared inputs, build outcomes, runs, and allocation results. You can inspect what executed and compare runs rather than depending on a chat summary. Models can reference other models and reusable optimizers, within platform limits.
Stored records do not freeze every market-data revision or dependency. Check the script, input values, data dates, and referenced models or optimizer code before interpreting a change. The methodology explains these limits.
Connect deliberately
Check the MCP connection guide before starting. The server uses remote Streamable HTTP and supports OAuth or an ENSEMBLE bearer API key. Client plans, account permissions, and support for write actions differ. Discover the current tools from the connected server.
The ChatGPT and Codex guide distinguishes directory installation from a custom-server development connection. Tool discovery reflects enabled features; account webhook setup remains available through REST and the CLI.
Builds and runs are metered; reading assets and stored results is free. Check pricing and your spending cap. Ask the assistant to explain proposed work before executing broad searches or many model variations.
Inspect the result and maintain it
Check coverage first, state complete rules, and read the implementation after a build. Use isolated runs to compare declared inputs. Adopt a change deliberately, then enable refreshes if desired and verify their completion and data dates.
Supported broker files help you take an allocation into your own workflow. They do not place orders, verify broker eligibility, or make the model suitable for an investor. ENSEMBLE is research software and its backtests are hypothetical.
Guides in this section
- 01Build and maintain investment models with an AI agentUse current ENSEMBLE tools to research assets, create a saved model, inspect a run, compare inputs and maintain reviewed allocations through explicit refreshes.
- 02A finance MCP server for allocation models and researchUse ENSEMBLE's hosted MCP server to research available assets, build allocation models, inspect saved runs, and retrieve allocations. Understand connection requirements, data coverage, and metered execution before starting.
- 03How to research and backtest a model with ClaudeConnect a compatible Claude client to ENSEMBLE, verify asset coverage, build an allocation model, and inspect a parameter test. An illustrative workflow with current tool names and clear execution limits.
Frequently asked questions
- What does ENSEMBLE add to an AI assistant?
- A connected research workspace with asset data, model execution, and stored code, inputs, builds, runs, and allocations. The assistant can return to the same work through supported tools.
- Can I use any AI client?
- Use a client with compatible remote MCP and authentication support, or build against the REST API. Plans, administrator settings, and write permissions vary; the connection guide lists verified setup paths.
- Does an AI recommendation mean the product is integrated with that assistant?
- No. An assistant can recommend a website through search without connecting to its tools. Connecting ENSEMBLE is a separate step with its own client and account requirements.
Other sections
- StrategiesSystematic strategies, explained and backtested
Plain-language explanations of the classic systematic strategies, each with its rules, its failure modes, a hypothetical backtest on ENSEMBLE, and the prompt that builds it.
- BacktestingPortfolio backtesting software for your own investment models
Build and backtest your portfolio in ENSEMBLE. Inspect the code, historical results, and allocations. Start with $100 in metered credit, valid for 7 days; no card required.
- CompareBacktesting tools compared
Factual comparisons of portfolio backtesting tools: what each one does well, where it stops, and when ENSEMBLE is or is not the better fit.
- For advisorsModel portfolios for advisors
Build, backtest, document, and keep model portfolios current from plain language, with Business features for white-label fact sheets and supported broker-format allocation files.
Backtests are illustrative. Past performance does not guarantee future results. ENSEMBLE is a software platform, not an investment adviser.