abliteration-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| DEBUG | No | Enable debug logging to stderr, e.g. `abl:*` for everything or `abl:http` for HTTP traffic only. | |
| ABLITERATION_API_KEY | Yes | Your abliteration.org API key, which looks like `abl_live_...`. Required; obtain one from https://abliteration.org/api. Can alternatively be passed as the command-line flag `--api-key`, which wins if both are set. | |
| ABLITERATION_API_BASE | No | Base URL for pointing the server at a local or staging server, e.g. `http://localhost:4321/api/v1` or the bare host `http://localhost:4321` (the `/api/v1` path is added automatically). |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| search_modelsA | Search the abliteration.org catalog of abliterated and uncensored LLMs (GET /models). Filter by M-code method, base family, author, license, download range and creation date. Delisted models are excluded unless delisted=dead|all. Cost: 5 credits up to 200 rows; 20 / 80 / 300 for larger pages, multiplied by 1 + (offset/2500)^2. Filter narrowly rather than paging deep. |
| get_modelA | Full record for one model (GET /models/{id}): method classification and evidence, family, size, files, quantizations, benchmarks, README, author summary. Delisted models are served with is_delisted=1 and days_since_gone. Cost: 10 credits. |
| list_methodsA | The M-code taxonomy of uncensoring methods (GET /methods): abliteration, weight editing, merges, fine-tunes, repackages, with model and download counts per code. Cost: 5 credits. |
| get_methodA | One M-code (GET /methods/{id}) with counts, top authors and sample models. Cost: 1 credit. |
| list_authorsA | Model publishers with aggregate model, download and like totals (GET /authors). Cost: 5-15 credits by rows returned. |
| get_authorA | One publisher with totals and their top models (GET /authors/{id}). Cost: 1 credit. |
| search_datasetsB | Alignment and uncensoring datasets (GET /datasets), filterable by workflow category, author and license. Cost: 5-15 credits by rows returned. |
| get_datasetA | One dataset with its files and passport, the plain-language summary (GET /datasets/{id}). Cost: 1 credit. |
| list_indicesA | All four flagship indices at once (GET /indices): DAI, Freedom Velocity, Weaponization and F2W Latency. Check |
| get_indexA | One index with its full history series (GET /indices/{id}). Only dai has history today. Cost: 10 credits. |
| get_timelineA | Curated field events, newest first (GET /timeline): delistings, takedowns, notable releases. Cost: 50 credits flat - the most expensive route; narrow with from/to. |
| list_delistedA | Change stream of models Hugging Face no longer serves (GET /models/delisted), preserved at their final state. Pass since = the newest last_seen_at from a previous call to get only new casualties. Cost: 5 credits flat. |
| get_deletions_summaryA | Aggregate of the 24-month Hugging Face deletion analysis (GET /models/historical-deletions/summary): total seen, total deleted, deletion rate, average lifespan, monthly losses, top-10 casualties. Cost: 20 credits. |
| list_deletionsA | Paged list of models deleted from Hugging Face over the last 24 months (GET /models/historical-deletions/list). Rows are frozen at last sighting. Cost: 5 credits flat. |
| list_readme_driftsA | Every commit that changed one model's README on Hugging Face (GET /models/{id}/drifts), with size_delta between commits. Pass since_ts = the max commit_ts from a previous call for an incremental feed; fetch a version with get_readme_at. Cost: 5 credits flat. |
| get_readme_atA | Wayback view: the exact README of one model at one commit sha (GET /models/{id}/readme-at/{commit}). Use a commit_sha from list_readme_drifts. Cost: 10 credits. |
| get_download_anomaliesA | Models whose downloads jumped (spike, >= 2x) or fell (drop, <= 0.5x) against a baseline window (GET /market/anomalies), biggest movers first. Cost: 15 credits flat. |
| get_download_historyA | Per-model download timeline (GET /models/{id}/download-history), the raw series get_download_anomalies scans. Cost: 5 credits flat. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 18 tools
Most tools have clearly distinct resource+action purposes (list/get/search pairings across authors, models, methods, datasets, indices). The main overlap is the deletion family (list_delisted vs list_deletions vs get_deletions_summary) and the download pair (get_download_anomalies vs get_download_history), but the descriptions differentiate them well enough for an agent to choose correctly.
Consistent verb_noun convention throughout: list_*, get_*, search_* across all 18 tools. Minor adjective-as-noun case in list_delisted, but it reads naturally and matches the pattern. No camelCase or style mixing.
18 tools is on the heavier side but each maps to a distinct route/resource of a genuinely broad data API (models, authors, methods, datasets, indices, timeline, deletions, README drifts, download anomalies). Nothing appears redundant, though it sits just above the ideal 3-15 range.
The surface covers a wide domain well: models, authors, methods, datasets, indices, timeline, delistings, deletions, README drift history, and download anomalies, with both list and detail variants. Minor gaps exist (e.g. no explicit search for authors/methods beyond list, no benchmark-only lookup), but core workflows are covered.