Signals Data MCP

Signals Data MCP turns the Signals disclosure corpus into a small, typed set of agent tools. It serves real public-company data with inspectable source URLs and verbatim supporting quotes over one remote endpoint:

https://api.signals.ai/api/infra/mcp

One public server, many clients

We do not maintain separate MCP servers for Codex, Claude Code, Cursor, or ChatGPT. Client-specific forks drift in names, schemas, citations, and access rules. Every external agent instead receives the same B2B tool contract and the same cited result for a given call.

There is a separate internal consumer-agent surface inside the Signals app. That surface can use user context such as watchlists or holdings and is governed by personalization rules. It is not exposed through the public MCP endpoint. Keeping it separate prevents a third-party B2B client from crossing into private user context or personalized portfolio behavior.

An eventual ChatGPT app or Claude directory connector should sit on top of this same public MCP server. If a rich MCP App UI is added, it should render results from these data tools rather than fork their retrieval logic.

What the server guarantees

• Read-only and idempotent tools; no orders, portfolio mutation, or webhooks.
• Stable JSON Schema inputs and declared output schemas.
• Discovery/fetch pairs so agents do not load full documents by accident.
• Provenance gating before data leaves the corpus.
• A structured JSON result plus a text fallback for older MCP clients.
• Exact tool names: anything absent from tools/list is rejected.

The server implements initialize, ping, tools/list, and tools/call using MCP protocol version 2025-06-18. It does not currently expose MCP resources, prompts, elicitation, or mutating tools.

Start with Connect a client, then use the tested workflows.