PaperPilot MCP Agentic Research Server
Small MCP demo project with:
- A basic MCP server (server.py) exposing a tool, resource, and prompt.
- A research server (research_server.py) that searches arXiv and stores results.
- Multiple client examples showing how to connect and use tools/resources/prompts.
Requirements
- Python 3.13+
uvinstalledOPENAI_API_KEYset for the OpenAI-backed clients
Install```sh
uv sync
## Run Servers
Basic demo server:```sh
uv run server.py
Research server:```sh
uv run research_server.py
## Client Examples
Minimal MCP client:```sh
uv run client.py
OpenAI tool-calling client (single server):```sh
uv run openai_mcp_client.py
Multi-server client with resources/prompts support:```sh
uv run prompt_resource_client.py
Multi-server client (tools only):sh
uv run remote_serverconnection_client.py
Configuration
server_config.json defines MCP servers used by the multi-server clients.
Default entries include:
- filesystem via @modelcontextprotocol/server-filesystem
- research via research_server.py
- fetch via mcp-server-fetch
You can add additional servers by extending server_config.json.
Notes
- The research server stores results in
papers//papers_info.json. - Set
MCP_TRANSPORT=streamable-httpto runserver.pyover HTTP transport.
