MCP (Model Context Protocol)
Model Context Protocol (Anthropic, 2024) defines a JSON-RPC-based interface where an MCP server exposes tools (callable functions), resources (readable data), and prompts (templated instructions) over stdio or SSE. Clients (LLM applications) discover these capabilities at connect time and invoke them as the model decides. The protocol is open; multiple language SDKs and a growing server ecosystem exist.
On This Page
Definition
MCP (Model Context Protocol)
Model Context Protocol (Anthropic, 2024) defines a JSON-RPC-based interface where an MCP server exposes tools (callable functions), resources (readable data), and prompts (templated instructions) over stdio or SSE. Clients (LLM applications) discover these capabilities at connect time and invoke them as the model decides. The protocol is open; multiple language SDKs and a growing server ecosystem exist.
Why it matters
MCP is becoming the de facto standard for agent-to-tool wiring. For trading agents this means a finance-MCP server (broker, market data, news) becomes the unified backend for any MCP-compatible client (Claude Desktop, custom agents, other LLMs). The same agent can swap between providers without re-implementing tool plumbing.
How it works
An MCP server registers tools (with JSON-Schema input/output), resources (URI-addressed data), and prompts (parameterized templates). On client connect, the server lists capabilities. The LLM client invokes tools by name with structured args; the server executes and returns structured results. Authentication, rate limits, and idempotency are server-side concerns; the protocol itself defines the wire format.
Example
Finance MCP server for an agent
Tools exposed
12 (get_quote, place_order, get_filings, ...)
Resources exposed
4 (positions, account, watchlist, news)
Auth
Bearer token from broker
Transport
stdio (local) + HTTPS+SSE (remote)
One server, any compliant client. Switch from Claude to GPT-5 to a custom agent without touching the trading-tool code.
Key Takeaways
MCP solves the N×M tool-integration problem (N agents × M tools) by standardizing the interface.
For trading agents, evaluate MCP servers on scope, auth, idempotency, transport, and schema discipline.
An MCP server with no idempotency guarantee on order placement is a production liability.
Related Terms
Try These Tools
Run the numbers next
Finance MCP Directory
Security-graded catalog of finance MCP servers — Alpaca, Polygon, Databento, IBKR, Tradier, Tiingo, NautilusTrader. Scope, auth, idempotency, transport.
Agent Skill Tester for Markets
Paste a SKILL.md definition + sample input + your Anthropic API key. See structured extraction, token cost, and latency — all in your browser. No signup.
FAQ
Questions people ask next
The short answers readers usually want after the first pass.
Sources & References
- Model Context Protocol specification — Anthropic / open standard
Related Content
Keep the topic connected
Agent Skill Testing
Agent skill testing: the regression-test discipline for LLM-driven agents. What to test, how to score, and the difference between pass-rate and capability.
Prompt Injection
Prompt injection: when untrusted text in a prompt overrides system instructions. The attack patterns and the structural defenses that work in production.