Communication kit for package discovery

This page is a reusable kit for spreading duckdb-vectors across both human and agentic audiences. Beyond automated discovery, we maintain a communication package and a list of referencing places for duckdb-vectors. For the full picture of what you are registering, start from the documentation home.

Automated discovery

Two discovery conventions are already served from the project itself, so most venues can point at them rather than restate the content:

• llms.txt — a curated, Markdown index for LLMs at the root of the documentation endpoint: llms.txt. The convention, proposed by Jeremy Howard in 2024, is an H1 project name, a one-line blockquote summary, then H2 sections of annotated Markdown links; it curates rather than dumps, and it does not control crawling. See llmstxt.org. We ship llms.txt only — llms-full.txt is deliberately out of scope.

• AGENTS.md — an agent-facing Markdown file sitting repository top-level: AGENTS.md, that can be used jointly by crawling agents and by agentic IDEs. It is the agreed "README for agents" standard that most coding agents now read. See agents.md.

Working a reference

Working on a new reference to duckdb-vectors consists of following steps:

1. Scope a suitable registry or reference where it would make sense to list duckdb-vectors
2. If this reference is already present in the table below, and tagged as complete, stop here. Maybe you would like to consider another reference target.
3. candidate → on-going. Open a pull request that updates this page, mentioning what you are working on, then do whatever is necessary to register duckdb-vectors at the referencing site. Read the venue's current submission rules at its link in the table below (they drift), then prepare the submission from the registration package. For the official MCP registry, the server manifest is already authored at the repo root — server.json (name io.github.bernard357/duckdb-vectors, PyPI package with the mcp stdio runtime). Re-validate it against the current $schema (the schema version drifts), then publish it with the mcp-publisher CLI, authenticating the io.github.bernard357 namespace via GitHub. For the Docker MCP Catalog, fork docker/mcp-registry, add the server definition, and open a pull request; every submission gets a Docker-team review before merge, and on approval Docker builds, signs, and publishes the image. For awesome-lists and framework directories, open a pull request adding one annotated link.
4. on-going. The submission is open and awaiting third-party review or merge. Record the pull-request or submission URL here next to the venue.
5. on-going → complete. Complete the pull request: update this page, then ask for a merge. The entry is live in the venue. Verify by loading the public listing, not by trusting the merge — confirm the entry resolves and the links in it work.

Keep this page honest: only mark a venue complete once its public listing is live and verified.

The registration package

Copy-paste these fields into any submission form, catalog entry, or pull request that helps referencing duckdb-vectors.

• Name: duckdb-vectors
• One-line pitch: A frugal, embeddable semantic-search building block for agentic systems — hybrid vector + exact-token search over a local DuckDB file, no external vector database.
• Description: duckdb-vectors indexes a modest, read-mostly corpus into a single DuckDB file and answers hybrid semantic + exact-token queries with a local function call in tens of milliseconds. The file bakes into the agent's own container image, so there is no external vector database to run, pay for, or keep alive. Embedding and chunking are pluggable; the surface is S3-Vectors-flavored. It also ships an MCP-over-stdio server so any MCP client can call its read-only retrieval tools.
• Keywords: duckdb, vectors, embeddings, semantic-search, rag, bm25, agents, mcp, llm, vector-search
• Install: pip install duckdb-vectors (add [bedrock] for the Amazon Titan embedder)
• MCP tool: duckdb-vectors mcp is an MCP-over-stdio server exposing two read-only tools — search(query, top_k=5) for hybrid retrieval and report() for index introspection. An MCP client spawns it as a child process and talks to it over stdin/stdout.
• License: MIT
• Links:
  • PyPI: https://pypi.org/project/duckdb-vectors/
  • Documentation: https://bernard357.github.io/duckdb-vectors/
  • Repository: https://github.com/bernard357/duckdb-vectors
  • Agent instructions: AGENTS.md
  • MCP manifest: server.json

External references

Each venue is a place where the package can be discovered. Use the candidate → on-going → complete lifecycle. Every candidate requires a manual maintainer or agent action, pending third-party review. Moving to complete requires an accepted review.

Venue	Type	Status
PyPI	Package metadata	complete
Official MCP registry	MCP server catalog	candidate
Docker MCP Catalog	MCP server catalog	candidate
awesome-mcp-servers	Curated awesome-list	candidate
awesome-rag	Curated awesome-list	candidate
LangChain integrations	Agent-framework tool directory	candidate
LlamaIndex integrations	Agent-framework tool directory	candidate

The candidates listed here are only suggestions. Please add any other reference you believe is worth the effort to spread duckdb-vectors.