Milestone 2 — MCP-Based Resume Matching System

Architecture Overview

┌─────────────────────────┐     JSON-RPC 2.0 (TCP)    ┌─────────────────────────────┐
│   matching_agent.py     │ ◄──────────────────────── │  filesystem_mcp_server.py   │
│   (LangGraph + GPT-4)   │ ──────────────────────── ►│  (MCP Server, port 8765)    │
│                         │                            │                             │
│  Nodes:                 │   tools/call → result      │  Tools exposed:             │
│  1. load_job_desc       │                            │  • read_file                │
│  2. load_resumes        │                            │  • write_file               │
│  3. watch_new_resumes   │                            │  • list_directory           │
│  4. match_candidates────┼──► GPT-4o (OpenAI API)    │  • search_files             │
│  5. save_report         │                            │  • get_file_info            │
└─────────────────────────┘                            │  • delete_file              │
                                                       │  • watch_directory ★        │
                                                       │  • batch_process ★          │
                                                       └─────────────────────────────┘

Setup

pip install -r requirements.txt
export OPENAI_API_KEY="sk-..."

Running

Step 1 — Start the MCP server (TCP mode)

python filesystem_mcp_server.py --transport tcp --port 8765

Step 2 — Run the agent (separate terminal)

python matching_agent.py --jd job_descriptions/senior_engineer.txt --resumes resumes/

Step 3 — Run tests

python -m pytest tests/ -v

Demo (all-in-one)

python demo_runner.py

Files

File	Purpose
filesystem_mcp_server.py	MCP server — JSON-RPC 2.0, 8 tools, stdio + TCP transport
mcp_client.py	Async MCP client used by the agent
matching_agent.py	LangGraph agent with 5 nodes, all I/O via MCP
tests/test_mcp_system.py	29 unit tests (JSON-RPC, tools, batch, watch)
demo_runner.py	End-to-end demo script
resumes/	Sample resume files (alice_chen.txt, bob_martinez.txt, priya_nair.txt)
job_descriptions/	Sample JD (senior_engineer.txt)
logs/	Server logs + generated match reports

MCP Protocol Details

The server implements MCP 2024-11-05 over JSON-RPC 2.0.

Lifecycle

Client → initialize (protocolVersion, capabilities)
Server → {protocolVersion, capabilities, serverInfo}
Client → initialized  (notification, no response)

Tool call

→ {"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"batch_process","arguments":{...}}}
← {"jsonrpc":"2.0","id":3,"result":{"content":[{"type":"text","text":"{...}"}],"isError":false}}

Error codes

Code	Meaning
-32700	Parse error (invalid JSON)
-32601	Method not found
-32602	Invalid params (missing required field)
-32001	File not found
-32002	Permission denied / path traversal

watch_directory

Polls a directory every N seconds for new *.txt files. Returns file_created events:

{"event": "file_created", "path": "resumes/new_candidate.txt", "timestamp": "...", "size_bytes": 1234}

batch_process

Processes multiple files in one RPC call. Operations:

• read_all — return full content of each file
• word_count — words, lines, chars per file
• extract_emails — regex-extracted emails per file
• summarize_stats — compact metadata for matching pipeline