Company Delta

What Company Delta MCP server does

The Company Delta MCP server is part of a repository containing two agent-facing interfaces. The Company Delta workflow accepts a company name and a starting date, then runs an analysis intended to identify material changes in the company. Results are returned as structured JSON, and the analysis uses an OpenAI-backed engine. The supplied service description also identifies primary sources as the basis for detecting changes in public companies.

The repository separately exposes a TSRE MCP tool for Swedish equity-market research. That tool returns the latest structured TSRE Daily Investment Desk, including market regime, breadth, sector rotation, RS20 candidates, Tactical Radar, data-quality details, and the underlying research contract. The feed is decision support only; it does not generate orders, autonomous investment advice, or buy/sell signals.

How it works

Company Delta uses A2A 1.0 rather than MCP for its company-analysis workflow. Clients can retrieve the Agent Card at /.well-known/agent-card.json, send JSON-RPC requests to /a2a/company-delta, and use SendMessage to create a tracked task. The request carries structured data with company and since fields. The service returns a task identifier when configured for immediate response, and GetTask is then used to poll for completion. Completed results appear as an application/json artifact.

Discovery and task polling do not start the analysis engine. The service limits A2A workload by default to one simultaneous job and three accepted jobs per client IP per hour. These limits can be changed with environment variables. An optional internal test token marks controlled verification traffic in structured logs without recording the token itself.

The TSRE interface is a separate MCP endpoint at https://machine-job-fishing-net.onrender.com/mcp/. It uses Streamable HTTP and provides one tool, get_sweden_market_intelligence, which takes no input and returns the current JSON feed.

Setup and configuration

For local development, create a Python virtual environment, install requirements-dev.txt, set OPENAI_API_KEY, run the test suite, and start the application with Uvicorn. The tests mock the costly OpenAI analysis, so passing tests do not perform a real Company Delta job. Keep the API key out of source control.

A2A concurrency and hourly acceptance limits use A2A_MAX_CONCURRENT_JOBS and A2A_RATE_LIMIT_PER_HOUR. Set A2A_INTERNAL_TEST_TOKEN only when controlled production verification requires internal request labeling, and send it in the X-A2A-Test-Token header.

Tools and capabilities

• Start a company-change task with A2A SendMessage.
• Poll a tracked task with A2A GetTask.
• Retrieve Swedish market intelligence with get_sweden_market_intelligence over Streamable HTTP.
• Inspect the MCP tool list before calling the TSRE tool.

Limitations and notes

A real Company Delta analysis requires OPENAI_API_KEY and may incur the cost associated with the OpenAI-backed engine. The A2A safeguards restrict concurrency and requests per client IP unless configured otherwise. The TSRE tool has no input parameters and returns the latest available feed rather than running a custom market query. Its output should not be treated as an order generator, autonomous investment advice, or a direct trading signal.