Index/Sheet F-01 · Search

Search

Semantic + full-text query UI over a repo’s graph — ask things like “how does authentication work here” or “what gotchas exist around database migrations” and get grounded answers.

01Four signals fused into one ranking: dense embeddings, BM25, exact-name matching, and personalized-PageRank expansion over the graph.

02Two-tier retrieval: a gist first, full detail only when the agent asks. Cheaper reads, same precision.

03Gotchas and decisions attached to a matched symbol come back in the same result.

Fig. F-01 — stylized view of the screen

agentops/Searchagentops · main · healthy

⌕how does read-only mode hide write toolsdense · bm25 · exact · ppr

All · 14Symbols · 9Gotchas · 1Decisions · 2Notes · 2

STRONGread_only_filteragentops-mcp/src/protocol.rs

Drops tools whose annotations mark them write-capable before the tool list is returned.

Decision attached: filter tools at list time, not call time

STRONGlist_toolsagentops-mcp/src/protocol.rs

Builds the tool list sent to the agent. Calls read_only_filter when the server is in read-only mode.

RELATEDToolRegistryagentops-mcp/src/protocol.rs

Reached by graph expansion from list_tools.

SUPPORTINGroute_mcpagentops-server/src/main.rs

Tier 1 · gist

Write-capable tools are removed from tools/list in read-only mode, so the agent is never told they exist. See read_only_filter.

Tier 2 · detail

Full source, callers and attached notes load only when the agent asks.

From the terminal

$ agentops search "how does authentication work here"

Agents get the same search over MCP; humans get this screen.

← 07Team & Settings 02 →Knowledge Graph

01 Search 02 Knowledge Graph 03 Documentation Viewer 04 Gotchas 05 Libraries 06 Repository Connection 07 Team & Settings