Lore
Give your coding agent the decisions your team already made — so it stops re-doing things you ruled out.
Your agent reintroduces an approach you rejected months ago. It rebuilds something you deliberately removed. The decision was written down — in an ADR nobody, human or agent, ever reopened.
Lore stores your requirements, decisions, designs, and roadmaps as typed Markdown in your repo, and serves them to Claude Code, Cursor, and Claude Desktop over MCP. The agent cites your decisions instead of violating them.
No AI in the core. No inference. No guessing. Just your team's recorded knowledge, in your Git, handed to the agent that needs it.
Lore is built on RAC — Requirements as Code — the open-source engine underneath. For now the package, CLI, and MCP server all ship under the rac name:
pip install requirements-as-code
📺 90-second demo — watch an agent violate a decision, then respect it. (link on launch)
Grounding your agent (start here)
Lore ships a read-only MCP server. Point your agent at your repo and it can search, retrieve, and traverse your recorded knowledge mid-task.
1. Install
pip install requirements-as-code
2. Connect your agent
Claude Code (from your repo root):
claude mcp add lore -- rac mcp
Claude Desktop / Cursor (mcpServers in the client config):
{
"mcpServers": {
"lore": { "command": "rac", "args": ["mcp", "--root", "/absolute/path/to/your/repo"] }
}
}
3. Ask, and watch it ground
"Should I add a hard delete to the user model?"
The agent calls Lore, finds your soft-delete decision, cites it by ID, and proposes the compliant change — instead of reintroducing the thing you removed on purpose.
The server exposes four read-only tools: get_artifact, search_artifacts, get_related, get_summary. It never writes to your repo.
▶ Full walkthrough + runnable example: examples/guide/
Why this works
The code is structured, the tests are automated, the infrastructure is versioned — but the reasoning behind what you build is scattered across tickets, chats, and dead docs. Agents can't act on what they can't read, so they re-litigate settled decisions.
Lore puts that reasoning back in the repo as typed, connected artifacts, then serves it to the agent through a deterministic interface. You write the decision once, in Markdown; RAC validates it, links it, and makes it retrievable — durable context for both humans and AI, with no proprietary format and no hosted platform.
Who it's for
- Teams running coding agents heavily (Claude Code, Cursor) who are tired of the agent ignoring decisions the team already made.
- Teams who already write ADRs and want those decisions to actually shape what the agent does.
- Anyone who wants the why behind their software versioned alongside the code.
How this relates to spec-driven development
Spec-driven development (SDD) tools — GitHub Spec Kit, OpenSpec, Kiro — manage the change: proposal, design, tasks, carried through to implementation. They treat requirements as ephemeral inputs that are consumed and archived. RAC manages the requirements: a durable, versioned, governed corpus that persists across changes and is served to your agent over MCP. RAC is the layer above SDD tools, not a competitor to them — an SDD tool drives each change, while Lore holds the decisions and requirements those changes draw on.
| Dimension | Lore / RAC | GitHub Spec Kit | OpenSpec |
|---|---|---|---|
| Requirement persistence | Requirements, decisions, designs, and roadmaps are long-lived artifacts that persist across changes | Spec, plan, and tasks are created per feature under specs/<feature>/ |
Change folders are archived on completion under openspec/changes/archive/; the specs directory is updated |
| Change management | None — RAC does not manage the change cycle; pair it with an SDD tool | Slash-command workflow: specify, clarify, plan, tasks, implement | Slash-command workflow: propose, apply, archive |
| Traceability | Typed Related links between artifacts; rac relationships --validate checks them in CI |
/speckit.analyze runs cross-artifact consistency and coverage analysis |
openspec validate checks changes and specs for structural issues |
| Tool coupling | Read-only MCP server; works with any MCP client | Slash commands or skills installed per agent at init (GitHub Copilot, Claude Code, Cursor, and others) | Slash commands for 20+ AI assistants |
| Install footprint | pip install requirements-as-code (Python 3.11+) |
uv tool install specify-cli from the Git repository (Python 3.11+) |
npm install -g @fission-ai/openspec (Node.js 20.19+) |
Authoring artifacts (the RAC CLI)
The MCP server is only as good as what it serves. RAC's CLI is how you write and maintain that knowledge — and how you enforce it in CI.
rac validate rac/ # check every artifact in a directory
rac inspect requirement.md # see its type and completeness
rac review rac/ # full repository review, worst problems first
New to Lore? Author your first artifact in five minutes: docs/quickstart.md.
Supported artifact types
- Requirements — what needs to exist
- Decisions — why choices were made (ADRs)
- Designs — product experience thinking
- Roadmaps — where the product is heading
- Prompts — reusable AI collaboration patterns
Everything stays plain Markdown — see docs/artifacts.md.
How Lore earns trust
Lore asks you to trust it with your product knowledge, so it holds itself to the same standard it applies to your repository:
- The MCP server is read-only by construction. It cannot create, modify, or delete files in your repo — enforced in code and verified by tests, not by convention.
- No AI in the core. Retrieval is deterministic: the same repo state and the same query always return the same result. The reasoning is your agent's job; Lore's job is to hand it the facts.
- It dogfoods itself. Lore's own planning corpus under
rac/is validated by RAC in CI — if the tool's rules break the tool's own artifacts, the build fails. - Output is a contract. Golden tests pin CLI and MCP output; any change to what the tools return is reviewed as a product change.
- Telemetry is opt-in, local-only, and content-free. Nothing is recorded without an explicit flag, events never include your arguments or repository content, and nothing leaves your machine unless you submit a report yourself — Lore contains no network code.
Documentation
- Quickstart — install and author your first artifact
- MCP server — tools, client configuration, examples
- CLI reference — every command, flag, and exit code
- Artifact types — the five types and their sections
- Relationships — link artifacts and validate the links
Requires Python 3.11+. uv tool install requirements-as-code also works.
Project status
Lore is early and evolving quickly. The MCP server ships today; feedback from teams running agents in anger is exactly what shapes what comes next. Contributions, ideas, and experiments welcome — see CONTRIBUTING.md.
License
MIT
Metadata
Release files for requirements-as-code 0.10.5
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| requirements_as_code-0.10.5.tar.gz | 2.3 MB | With provenance | Details for requirements_as_code-0.10.5.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| requirements_as_code-0.10.5-py3-none-any.whl | Python 3 | none | any | Details for requirements_as_code-0.10.5-py3-none-any.whl |
Total release size: 2.5 MB
Release files / requirements_as_code-0.10.5.tar.gz
| Download URL | requirements_as_code-0.10.5.tar.gz |
|---|---|
| Size | 2.3 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
ba16d57a5eac20691201f5b27cbfaef1a78fa505bd1f44fffeb215fb831d3d02
|
|
BLAKE2b-256 checksum How to use checksums |
207d0774a51da1c14c111e097d52808263b9042fc0654df9da6ed0deb2019f18
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.13.12
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Jun 12, 2026.
Transparency logRelease files / requirements_as_code-0.10.5-py3-none-any.whl
| Download URL | requirements_as_code-0.10.5-py3-none-any.whl |
|---|---|
| Size | 183.9 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
ef34058eb72e62c9732371f5865f7b6f9767eed7e0e9b470a15a5828083cffa5
|
|
BLAKE2b-256 checksum How to use checksums |
4146766963008de9b16f9f283deea3658a9007f452aaf020c10668654e4077c3
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.13.12
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Jun 12, 2026.
Transparency logSecurity for requirements-as-code 0.10.5
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
Full provenance
Both files in this release have been uploaded with provenance metadata.
Attestation for requirements_as_code-0.10.5.tar.gz
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Jun 12, 2026.
Transparency logA note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.