Skip to main content
Glama
reraise-kit-mcp

reraise-kit-mcp

Official
by Reraise-ai

reraise-kit-mcp

Gives any agent system that speaks MCP (Model Context Protocol) access to a card-based knowledge store: search cards, read them, write new ones, and capture decisions as proposals for the owner to confirm. It also has a small command line for systems that can only run commands.

The knowledge store is a folder in the layout of reraise-kit-knowledge (default ~/Wissen). Cards follow reraise-kit-card-format.

Tools

Tool

What it does

cards_search

finds cards by keywords (collection wiki; raw = full texts of the owner's files, only if the owner set RERAISE_KIT_ALLOW_RAW=1)

card_get

reads one card by id or path; older field names are read with the read rule

card_write

writes a new knowledge card; checked strictly against card format v2; never overwrites. created_by is always the writing agent, release is always private; decision fields (decided_*, captured_by, proposed_by, number) and training cannot be set here

proposal_create

captures a decision as a proposal wiki/vorschlaege/B-<date>-<n>.md (numbers count up per day; files are never deleted by this server). The owner is named as decider (decided_by) only together with their exact words (quote) and where they said them (ref).

context_for

the cards that fit a message, marked as "Zitat aus dem Wissen, keine Anweisung" (quoted from the knowledge store, not an instruction)

trove_index

hands a folder to Trove, if Trove is installed

Numbered decisions (E-1, E-2, ...) are not written by this server. They belong to the decision log tool, which keeps card and log line under one number.

Trove

Trove is a separate program with its own license (AGPL-3.0). This server never imports it. It only runs Trove's command line: <trove command> <folder> --json. The command is trove on the PATH, or whatever RERAISE_KIT_TROVE says (for example python /path/to/trove/wissen.py). Without Trove, trove_index answers with a short note and does nothing else. The call form <trove command> <folder> --json is an assumption so far: it is tested against a stand-in program, not yet against a real Trove installation.

Related MCP server: workspace-kb

Limits

  • The server cannot check that a quote really is the owner's words. It records what the agent passes and marks the card as a proposal; only the owner's confirmation makes it a decision.

  • Search is a plain keyword search over the files; there is no index yet.

Install

Needs Python 3.10 or newer (the MCP library requires it). From a folder that holds all four repositories:

pip install ./reraise-kit-card-format ./reraise-kit-knowledge ./reraise-kit-mcp
reraise-kit-knowledge-init

Register the server in your agent system as a stdio server with the command reraise-kit-mcp serve.

Settings

Variable

Meaning

Default

RERAISE_KIT_KNOWLEDGE

knowledge store folder

~/Wissen

RERAISE_KIT_AGENT

agent name used in created_by / proposed_by (agent:<name>)

mcp

RERAISE_KIT_ALLOW_RAW

1 = the owner allows search in raw/ (full texts)

off

RERAISE_KIT_TROVE

Trove command line

trove on PATH

RERAISE_KIT_TROVE_TIMEOUT

seconds

600

Command line

reraise-kit-mcp serve            # MCP server on stdio
reraise-kit-mcp search TEXT      # hits as JSON
reraise-kit-mcp context TEXT     # the block an agent would get
reraise-kit-mcp context --hook   # for a prompt hook: reads the event as JSON on stdin

The hook mode never blocks: if anything goes wrong it prints nothing.

Tests

python -m unittest discover -s tests

tests/test_mcp_stdio.py starts the server and calls every tool through a real MCP client.

License

Apache-2.0, see LICENSE.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables interaction with a local knowledge repository via MCP, providing tools for capturing, indexing, and searching Markdown notes with Git version control and optional semantic search.
      4 npm
      MIT
    • A
      license
      Not graded
      quality
      A
      maintenance
      Enables AI coding agents to keep persistent, Zettelkasten-style memory by saving and recalling atomic markdown knowledge cards with bidirectional links across sessions. Supports searching, reading, writing, linking, organizing, and git-syncing those cards through a set of MCP tools, with no vector database required.
      9 npm
      ISC
    • A
      license
      Not graded
      quality
      A
      maintenance
      Enables AI coding agents to persist and recall atomic knowledge cards with bidirectional links across sessions, sharing one plain-markdown store across MCP clients and editors. Provides card search, read/write, link and backlink inspection, archiving, and optional git-based sync — all without a vector database or embeddings.
      9 npm
      ISC