# precis-mcp [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/retospect/precis-mcp  
**GitHub Stars:** 8  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/precis-mcp

## Description
MCP server giving LLM agents a seven-verb API over papers, documents, code, state, patents, and cached web/Wolfram/YouTube tool calls

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "precis-mcp": {
    "command": "uvx",
    "args": ["precis-mcp"]
  }
}
```

## Documentation

## What precis-mcp does

precis-mcp provides a single MCP tool surface for research, writing, technical design, agent work, memory, evidence, and selected external lookups. It can work with ingested papers, patents, proposal documents, conversations, slide decks, skills, and files under `PRECIS_ROOT`. Authored content includes chunk-based drafts, analytic CAD solids, structural-envelope designs, molecular-machine designs, atomistic structures, PCB graphs, and folders.

The server also models work and evidence as persistent objects. Quests represent open problems that can be advanced over repeated ticks; todos can recur and create work; jobs represent execution units; and plans, defect reports, memories, citations, findings, provenance, and agent logs provide supporting records. Tool kinds cover local SymPy calculations, Wolfram queries, YouTube transcripts, web extraction, Wikipedia, web search, and Perplexity research modes.

## How it works

Operations are selected through a verb and a `kind` value. Use `get` when you know an identifier or need to invoke a tool, `search` for topical or phrase-based discovery, and `put` to create a reference. `edit` changes file regions using content anchors, while `delete`, `tag`, and `link` manage existing records. Truncated results can be continued with `more` and its cursor.

Reference records may be addressed by slug or numeric ID. Responses provide compact handles such as `pa5` for a paper or `me42` for a memory. Content searches combine PostgreSQL full-text indexing with pgvector semantic search and reciprocal-rank fusion. File regions use a compact selector form such as `notes--meeting~L42-58`.

## Setup and configuration

Install the package with `uv` using Python 3.13:

```bash
uv tool install --python 3.13 'precis-mcp[all]'
```

The installation creates an isolated environment containing the `precis` command. Upgrade it with `uv tool upgrade precis-mcp`. The core installation is approximately 650 MB, while the all-extras installation is approximately 1.6 GB on the documented macOS arm64 measurement; Linux CUDA dependencies can require several additional gigabytes. Models are downloaded on first use.

Optional extras and environment configuration determine which kinds appear in a running instance. The live registry can be inspected with `get(kind='skill', id='precis-help')`; `precis-overview` provides a guided tour, and `precis-toolpath-help` maps goals to calls. Single-machine and multi-host deployment paths are included in the project documentation.

## Tools and capabilities

- Search papers, patents, documents, code, memory, and other indexed content.
- Preserve claims with paragraph-level citations and evidence chains.
- Create drafts that export to LaTeX, PDF, and Word, with figures, diagrams, and audio narration.
- Navigate Python repositories using symbol and call-graph information.
- Build and analyze CAD, molecular, atomistic, and PCB representations.
- Manage quests, todos, jobs, plans, agent fleets, links, tags, and provenance.
- Query cached web, Wolfram, YouTube, Wikipedia, web search, and Perplexity sources when enabled.

## Limitations and notes

The available kind set depends on installed optional dependencies and configured environment values. A capability whose requirements are unavailable is omitted from the tool surface. The catalogue is therefore build-specific; the live registry is the authoritative way to discover enabled kinds. Some operations also depend on configured services such as PostgreSQL, pgvector, or external research tools. The project is actively developed, and its catalogue changes over time.

_Full upstream README: https://allmcps.com/mcp/precis-mcp/readme_