ucp-gen
Generate Universal Context Packages from
real systems. Two sources are supported — GitHub issues and Jira tickets —
and one command turns an issue with its comments, history and links into a
validated, provenance-backed .ucp.json. By default no LLM is involved:
the structure alone already carries the facts, decisions and timeline.
pip install ucp-gen
# GitHub (GITHUB_TOKEN optional, raises the API rate limit)
ucp-gen github vercel/next.js#12345 -o task.ucp.json
# Jira (Cloud: email + API token; Server/DC: personal access token)
export JIRA_BASE_URL=https://yourco.atlassian.net
export JIRA_EMAIL=you@yourco.com
export JIRA_API_TOKEN=...
ucp-gen jira PROJ-123 -o task.ucp.json
# canonical LLM rendering, capped at 1500 tokens
ucp-gen github owner/repo#42 --markdown --token-budget 1500
# include a "what changed since" diff (adds the ucp-temporal profile)
ucp-gen github owner/repo#42 --since 2026-06-01T00:00:00Z
# pretty-print any package in the terminal
ucp-gen view task.ucp.json
The CLI is built for humans: spinners while fetching, checkmarks per step,
a summary tree after writing, rich --help with grouped options, and an
interactive prompt when you omit the issue reference. Decorations go to
stderr — stdout stays pure JSON/Markdown, so piping is always safe:
✓ pallets/flask#5961 — issue + 4 comments + 1 linked PRs
✓ valid ucp-core package — 6 sources, sha256-hashed
📦 wrote task.ucp.json
├── Flask 3.1.3 test breaks after Werkzeug update…
├── claims 7 must-know
├── decisions 1 (1 accepted)
├── sources 6
└── tokens ~713 rendered
Coverage and decision dedup (0.3.1+)
Every package includes a coverage block: whether material was truncated,
how many upstream artifacts were considered vs included, and per-stream counts
(comments retrieved vs represented, timeline fetch limits). On mega-threads
like microsoft/vscode#519 you get truncated: true with available: 596
even when only 200 comments were fetched.
When a merged linked PR exists, proposed decisions extracted from comment phrases like "we decided …" are dropped — the merged PR is the authoritative accepted signal (SPEC §4.11).
Optional LLM enhancement (--llm)
Structure tells you what happened; it cannot tell you which of 200
comments contains the key insight. --llm adds that layer with a single
call to any OpenAI-compatible endpoint (OpenAI, kie.ai, OpenRouter, a
LiteLLM proxy, local Ollama):
export UCP_LLM_BASE_URL=https://your-provider.example/v1 # any OpenAI-compatible URL
export UCP_LLM_API_KEY=...
export UCP_LLM_MODEL=your-model
ucp-gen github owner/repo#42 --llm -o task.ucp.json
What it changes: summary becomes a real synthesis of the whole thread
(marked with confidence), comments the model flags as important get a
salience boost, and decisions/conflicts stated in prose are extracted.
Provenance survives: the model may only cite the source keys it was given —
hallucinated citations are dropped, and every added claim still points at a
real, hashed source. The model used is recorded in generator.llm_model.
If the call fails, you get the structure-only package and a warning, never
a broken one.
A real run on microsoft/vscode#519 (596 comments over ten years; 200 in
the package, 17 sources, ~1,623 rendered tokens) shows the difference: the
enriched summary explains why the feature was never built, a conflict
captures the "Electron is at fault" vs "VS Code's hard-coded styles are at
fault" dispute with both positions citing hashed comments, and a decision
with status rejected records that the request is off the roadmap — none
of which exists in any structured GitHub field.
What the mapping does
| GitHub | Jira | UCP |
|---|---|---|
| title / state / assignee | summary / status / assignee | entity |
| first meaningful paragraph | first meaningful paragraph | summary |
| state, milestone, labels, PR states, comment gists | status+resolution, priority, due date, fix versions, links, comment gists | must_know claims with salience |
| merged linked PRs | resolution | decisions (accepted; supersedes proposed comment decisions) |
| "we decided" comments | "we decided" comments | decisions (proposed, unless merged PR exists) |
| issue timeline | changelog | history, and context_diff with --since |
| fetch limits / comment counts | — | coverage (truncated honesty) |
| — | "is blocked by" links | dependencies |
| linked PRs | links, parent, subtasks | related_objects |
| every cited issue / comment / PR | every cited ticket / comment | sources with URL + content hash |
Every claim cites its sources; every source carries a sha256 content hash.
The output always validates against the UCP schema before it is written —
the generator will fail rather than emit an invalid package.
Feed the result to any UCP consumer, e.g. the reference MCP server:
pip install ucp-mcp
ucp-gen github owner/repo#42 -o ./contexts/task.ucp.json
ucp-mcp --dir ./contexts # Cursor / Claude Code now sees the context
Development
pip install -e ".[dev]"
pytest
Part of the UCP reference toolchain (Apache 2.0).
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file ucp_gen-0.3.1.tar.gz.
File metadata
- Download URL: ucp_gen-0.3.1.tar.gz
- Upload date:
- Size: 23.2 kB
- Tags: Source
- Uploaded using Trusted Publishing? Yes
- Uploaded via:
twine/6.1.0 CPython/3.13.12
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
ec1804e943c80397ed1670ced9e07171898e12891bb7110a897551102b3cbf78
|
|
| MD5 |
f24977fbf61be66af6e7a84da97aebf4
|
|
| BLAKE2b-256 |
b080ce6ea7563a200c8f29edb42a903212529b305e1ec2674cd78493e746408c
|
Provenance
The following attestation bundles were made for ucp_gen-0.3.1.tar.gz:
Publisher:
release.yml on ucpcore/ucp
-
Statement:
-
Statement type:
https://in-toto.io/Statement/v1 -
Predicate type:
https://docs.pypi.org/attestations/publish/v1 -
Subject name:
ucp_gen-0.3.1.tar.gz -
Subject digest:
ec1804e943c80397ed1670ced9e07171898e12891bb7110a897551102b3cbf78 - Sigstore transparency entry: 2084370808
- Sigstore integration time:
-
Permalink:
ucpcore/ucp@9c3ec0a35e753a023ab4fa620558b183789b4f33 -
Branch / Tag:
refs/tags/gen-v0.3.1 - Owner: https://github.com/ucpcore
-
Access:
public
-
Token Issuer:
https://token.actions.githubusercontent.com -
Runner Environment:
github-hosted -
Publication workflow:
release.yml@9c3ec0a35e753a023ab4fa620558b183789b4f33 -
Trigger Event:
push
-
Statement type:
File details
Details for the file ucp_gen-0.3.1-py3-none-any.whl.
File metadata
- Download URL: ucp_gen-0.3.1-py3-none-any.whl
- Upload date:
- Size: 22.9 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? Yes
- Uploaded via:
twine/6.1.0 CPython/3.13.12
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
c2ab527a8b1cb562c91cbc2047758248717ae3260c144af71961eee5df3c04a2
|
|
| MD5 |
52e80360214258f46e18d52ce818e362
|
|
| BLAKE2b-256 |
d03dac085ee74f445f082b25b0e55d3b1c9844cb0c1e2e6b6e2c0986bbe275dc
|
Provenance
The following attestation bundles were made for ucp_gen-0.3.1-py3-none-any.whl:
Publisher:
release.yml on ucpcore/ucp
-
Statement:
-
Statement type:
https://in-toto.io/Statement/v1 -
Predicate type:
https://docs.pypi.org/attestations/publish/v1 -
Subject name:
ucp_gen-0.3.1-py3-none-any.whl -
Subject digest:
c2ab527a8b1cb562c91cbc2047758248717ae3260c144af71961eee5df3c04a2 - Sigstore transparency entry: 2084370830
- Sigstore integration time:
-
Permalink:
ucpcore/ucp@9c3ec0a35e753a023ab4fa620558b183789b4f33 -
Branch / Tag:
refs/tags/gen-v0.3.1 - Owner: https://github.com/ucpcore
-
Access:
public
-
Token Issuer:
https://token.actions.githubusercontent.com -
Runner Environment:
github-hosted -
Publication workflow:
release.yml@9c3ec0a35e753a023ab4fa620558b183789b4f33 -
Trigger Event:
push
-
Statement type: