Using with agents
RepoSniffer is primarily an MCP server for coding agents. It gives agents a hallucination-free way to find and verify open-source projects before adopting them.
What the agent gets
| Tool | Purpose |
|---|---|
find_repos | Feature query → ranked candidates with score breakdown, snippet evidence, license verdict, recommendation |
repo_intel | Verify an existing repo (alive? licensed? best-of-kind?) + 2 alternatives |
health | Embedding backend, model, GitHub auth status |
Every result carries an as_of freshness timestamp, a flags list
(archived, no-license, strong-copyleft, stale, …), a license_category
(permissive / weak-copyleft / strong-copyleft / unknown) and a targeted
snippet showing why the repo matched.
Installation
uvx --from reposniffer reposniffer-mcpThe server speaks MCP over stdio. The first tool call warms up the embedding model in the background.
opencode
Add to your opencode.json:
{ "mcp": { "reposniffer": { "type": "local", "command": ["uvx", "--from", "reposniffer", "reposniffer-mcp"], "enabled": true, "environment": { "GITHUB_TOKEN": "ghp_..." } } }}If the first call times out, raise the MCP timeout:
{ "experimental": { "mcp_timeout": 120000 } }Restart opencode after adding the server.
Claude Code
claude mcp add reposniffer -- uvx --from reposniffer reposniffer-mcpCodex and Cursor
Add an MCP server pointing at uvx --from reposniffer reposniffer-mcp (stdio) in the
respective MCP configuration UI.
Agent workflow example
- The agent needs a library for a task it hasn’t built before.
- It calls
find_repos(query="websocket rate limiting", intent="adopt"). - It reads the top result’s evidence snippet and recommendation.
- Before committing, it calls
repo_intel(owner_repo="laurentS/slowapi")to confirm the repo is alive, licensed, and best-of-kind — instead of hallucinating. - It cites the
as_oftimestamp in its answer.