Long-term memory for your coding agents.

Claude Code, Cursor, and Cline forget everything the moment you /clear. dropdat keeps the transcript, indexes it, and serves it back to the next session through one MCP tool surface — shared across every agent you use.

The MCP surface

dropdat_recall

Hybrid semantic + keyword search. Pulls the right capsules even when the agent asks in totally different language.

dropdat_read

Fetch one capsule’s full contents — every message, every version, optional lineage.

dropdat_list

Browse recent capsules, optionally filtered by tag.

dropdat_capsule

Save the current conversation slice as a new capsule.

dropdat_autocapsule

Save the full verbatim Claude Code session by reading the on-disk .jsonl transcript.

Install in your agent

FAQ

Why do coding agents lose memory?

Models have a finite context window. Once a session ends — /clear, /compact, IDE restart — the in-memory transcript is gone unless something external stored it. MCP gives agents a tool surface; dropdat is the backend that surface talks to.

How is this different from project-level CLAUDE.md files?

CLAUDE.md is static, hand-curated, and loads on every run. dropdat capsules are dynamic, searchable, and shared across all your agents — Claude Code can recall a capsule that Cursor saved last week.

Does dropdat_autocapsule grab the whole session verbatim?

Yes. It reads Claude Code's on-disk .jsonl transcript directly, so the capsule is byte-perfect — every message, every tool call, every system reminder.

What does semantic recall mean here?

Queries embed via OpenAI, run against pgvector cosine similarity, and fuse with BM25 keyword ranking. Ask 'how did we handle the rate-limit retry?' and matches surface even if the original capsule used different words.