Limitless Library
MCP server for verified reuse infrastructure that lets agents query before work to obtain safe, permissioned results (exact component, source-free method, or abstention) with receiver-owned verification and non-disclosure guarantees.
README
<p align="center"> <img src="docs/assets/limitless-library-mark.png" width="148" alt="Limitless Library mark"> </p>
<h1 align="center">Limitless Library</h1>
<p align="center"><strong>Verified reuse infrastructure for agents.</strong></p>
<p align="center"> <a href="LICENSE"><img alt="Apache-2.0 license" src="https://img.shields.io/badge/license-Apache--2.0-111111"></a> <a href="https://github.com/Univeracity/limitlesslibrary/actions/workflows/ci.yml"><img alt="CI status" src="https://github.com/Univeracity/limitlesslibrary/actions/workflows/ci.yml/badge.svg"></a> <img alt="Python 3.11 or newer" src="https://img.shields.io/badge/python-3.11%2B-111111"> <a href="docs/PROTOCOL.md"><img alt="MCP 2026-07-28" src="https://img.shields.io/badge/MCP-2026--07--28-111111"></a> <img alt="Pre-alpha" src="https://img.shields.io/badge/status-pre--alpha-6b7280"> </p>
<p align="center"> <a href="#quick-start">Quick start</a> · <a href="#how-verified-reuse-works">Lifecycle</a> · <a href="#mcp">MCP</a> · <a href="#documentation">Documentation</a> · <a href="CONTRIBUTING.md">Contributing</a> </p>
Reuse work without giving up trust
AI agents repeatedly solve problems that another agent has already solved. That wastes time and compute. Search alone does not make an earlier result safe to reuse: its authorization, compatibility, integrity, and actual adoption still need proof.
Limitless Library lets agents check before starting from scratch. It returns one permissioned result that fits the receiving environment—an exact component or source-free method—or abstains without disclosing a candidate. The receiver decides what may enter, verifies it locally, and records evidence that adopted work was actually used.
At a glance
| Property | What it means |
|---|---|
| Three safe outcomes | Exact component, source-free method, or abstention |
| Receiver-owned proof | The receiving environment chooses installation paths and verification programs |
| Exact bytes | Content-addressed installation refuses overwrite and rolls back on failure |
| Fail-closed verification | No network, inherited secrets, or unsandboxed verifier fallback |
| Observed use | Delivery is not counted as reuse until receiver code invokes the supplied component |
| Local by default | The reference lifecycle needs no account, hosted service, or model API |
Quick start
Try the complete local lifecycle from a source checkout. You need Python 3.11 or newer and Bubblewrap on Linux. On Debian or Ubuntu, install the one system dependency if it is not already present:
sudo apt-get install bubblewrap
Then run:
git clone https://github.com/Univeracity/limitlesslibrary.git
cd limitlesslibrary
./scripts/limitless
That is the only project command required. On first use, the launcher creates
an isolated environment under .limitless/, installs the small Python
dependency set, runs exact adoption, method guidance, and safe abstention, and
retains inspectable evidence under .limitless/quickstart. It requires no
environment activation, account, cloud service, model API, or model download.
The initial dependency installation may require package-index access; the
lifecycle itself runs without network access.
Check host readiness or retain another run at a path you choose:
./scripts/limitless doctor
./scripts/limitless demo --workspace ./limitless-demo
The demo refuses to overwrite an existing workspace. Later launcher runs reuse the isolated environment; running it without arguments keeps the original quick-start evidence and performs a disposable replay.
When integrating Limitless into another environment, install the checkout with
python -m pip install .; the installed commands are limitless and
limitless-mcp.
What the demo does
The bundled demo performs useful work on a structured agent audit event. It installs a narrow field-name redaction component and returns a sanitized copy while preserving non-sensitive data. It then demonstrates every safe outcome:
- Exact adoption — installs exact bytes without overwrite, runs receiver-owned checks, and proves receiver code invoked the component.
- Method guidance — returns a source-free method when Python bytes do not fit a JavaScript receiver.
- Safe abstention — withholds candidate details when no result fits.
The redactor protects configured structured field names; it is not a general scanner for secrets embedded in free text. See the local demo evidence map for the generated artifacts and manual inspection commands.
How verified reuse works
For exact reuse, the receiver—not the capsule—controls the trust boundary:
- A query describes the requested capability and receiving environment.
- Policy and compatibility gates select at most one eligible result.
- The receiver chooses the destination and provides digest-bound checks.
- Limitless installs content-addressed bytes without overwriting receiver files.
- Bubblewrap runs the checks with no network, no inherited secrets, bounded resources, and a read-only receiver mount.
- Runtime evidence proves invocation and an append-only receipt binds the decision, component, verification, and disposition.
If exact bytes do not fit, only a source-free method may cross. If no safe candidate exists, Limitless abstains without disclosing one.
What is open here
- JSON Schemas for capsules, queries, decisions, receiver recipes, verifier results, and adoption receipts.
- A source-minimized local catalog with policy and compatibility checks.
- Exact-byte, no-overwrite installation with rollback on failure.
- Receiver-owned adherence and obligation verification under Bubblewrap.
- Python embedding and bounded stdio MCP connector surfaces.
- Authoring and sealing commands, a one-command demo, conformance fixtures, and tests.
This repository is the inspectable foundation of the Limitless product. Its trust and foundation commitments are described in Open-source foundation.
Manual conformance proof
The smaller examples/ fixture exposes each primitive separately:
limitless validate-catalog --catalog examples/catalog
limitless query \
--catalog examples/catalog \
--request examples/requests/exact-python.json
Continue through installation, method guidance, abstention, and authoring in the conformance example.
MCP
Start the bounded local stdio server:
limitless-mcp --catalog examples/catalog
It exposes limitless_query_before_work and supports stateless MCP
2026-07-28 requests plus the legacy 2025-03-26 initialization flow. MCP is
a decision channel, not an artifact transport; installation remains an
explicit receiver-local operation.
Python callers can use query_local(...) or McpStdioConnector. See
Protocol.
Authoring
The checked-in examples/authoring records omit derived digests. Seal them
only after payload and receiver-verifier review:
limitless seal-capsule \
--draft examples/authoring/capsule.draft.json \
--root examples/catalog/hello-component \
--output capsule.json
limitless seal-recipe \
--draft examples/authoring/recipe.draft.json \
--receiver examples/receiver \
--output recipe.json
Outputs are immutable: both commands refuse to overwrite an existing path.
Documentation
| Guide | Purpose |
|---|---|
| Local demo | Evidence map and manual inspection |
| Protocol | Local API and MCP contract |
| Trust model | Security boundaries and failure behavior |
| Conformance | Required outcomes and fixture expectations |
| Open-source foundation | Relationship between the open implementation and product |
| Examples | Primitive-by-primitive walkthrough |
Status and security
This is a pre-alpha reference implementation. It is intentionally local, single-operator, and Linux-only for contained verification. It is not yet a remote trust system or package manager, and technical integration is not treated as product adoption.
There is deliberately no unsandboxed verifier fallback. Querying and method
selection work without Bubblewrap; exact adoption fails closed if containment
is unavailable. Schemas and interfaces may change incompatibly before 0.1.0
stable.
Read the trust model before use. Report vulnerabilities according to SECURITY.md.
Contributing and license
Contributions should preserve fail-closed behavior, non-disclosure on abstention, and receiver ownership of verification. Start with CONTRIBUTING.md and the Code of Conduct.
Apache-2.0 licensed. See LICENSE and NOTICE.
<p align="center"> A <a href="https://univeracity.com">Univeracity</a> project · <a href="https://limitlesslibrary.com">limitlesslibrary.com</a> </p>
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.