Phaedrus — Claude Code MCP server: research feeds and findings
A server that connects Claude Code to a store of research findings, tracked literature sources, and tools for analyzing and submitting research candidates. Works with local files, no database required.
Open on GitHubWhat Phaedrus does
Phaedrus is a Rust binary that speaks MCP over stdio to Claude Code and exposes seven tools for managing a research feed. The capability_context tool surfaces the capability registry from a paired Lupin server (used for planning); feed_summary gives counts of findings by system and type. The finding_get and finding_list tools retrieve curated findings; research_submit files a candidate (finding, roadmap item, or test) into a scan inbox without modifying the curated store. The server connects two data directories: one for findings and watchlist, another for read-only access to Lupin's capability context (defaults are ~/workspace/research-feed-20261019 and ~/workspace/revenue-alignment-20261019).
Writes are intentionally constrained: revenue_candidate_draft only assembles a draft payload for Lupin's opportunity submission, never writing to Lupin's own files. The server uses NDJSON JSON-RPC and has no network listener.
What it helps with
- Track curated research findings and summarize counts by system
- Submit research candidates to a scan inbox for review
- Query the list of monitored literature venues and search terms
- Draft a revenue opportunity submission from a research finding
Example prompts for the agent
- List all findings for the Deep Learning system
- Submit a candidate finding: novel gradient optimization approach, potential for revenue
- Show me the watchlist and how many findings we have per system
How to install Phaedrus
Install
The MCP registry has no launch command for it: see the repository README
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about Phaedrus
How do I install Phaedrus in Claude Code?
Download the Linux x64 binary from the Phaedrus GitHub releases. Add an MCP server entry in Claude Code's configuration (typically ~/.claude/config.json or via the MCP UI): point command to the binary path, pass --root and --lupin-root in args if you use custom directories. Reload Claude Code.
What does Phaedrus do?
It is an MCP server for a research feed: it stores curated findings, maintains a watchlist of literature sources, exposes Lupin's capability registry for planning, and lets the agent submit research candidates for review. All data lives in local files.
Does Phaedrus require a database?
No. The server operates on plain files in two directories: one for findings and the watchlist, another for reading capability context from Lupin. No network connections or external databases.
