Callosium MCP server

Read from the package source without running it. The installed server may list more.

• append_note

  ADDITIVE edit: appends content to a note (under a specific heading if given, else at the end). Prefer this over write_note for adding facts to an existing note — it can never destroy what is already there.

• archive_note

  Retire a note from recall/search WITHOUT deleting it: sets status: archived with your attribution and reason. Reversible by the owner. Agents can never hard-delete anything.

• brain_check

  Validate the whole brain against its schema: broken links, orphans, frontmatter problems, sync conflicts. Report only, never destructive.

• fetch_document

  Return the FULL text of a note, or of EVERY note under a folder, concatenated in path order — for rewriting/editing a complete document that may span several files. Pass a note path for one whole note, or a folder path for the whole folder. Large results are capped: notes past the character budget are listed (path + size) instead of inlined, so fetch a sub-folder for those. Use this over read_note

• gather

  Assemble a CONTEXT PACK for a topic/project: the most relevant notes with their PATHS, dates, and a real excerpt of each — so you get broad context in ONE call, then read_note / fetch_document the ones you need in full. Use for "brief me on X", "what's the status of / pending on X", "get me up to speed on X". Broader than recall (a focused answer) and richer than search (thin snippets).

• get_filing_rules

  The rules for turning raw material (documents, exports, notes) into filed notes in THIS brain: the partitions and what belongs in each, the routing order, naming, required frontmatter, and the ground-truth protocol for verbatim sources. Call this before filing anything so every note lands in its right home and the map stays true. YOU do the reading and distilling; these rules tell you where each n

• get_instructions

  The operating instructions for agents connected to this brain — replaces per-tool CLAUDE.md files.

• get_map

  The routing MAP of this brain — how it is organized and how to navigate it: the top-level structure (what lives in each folder), the topic hubs to start from, and how to find things. READ THIS FIRST when you connect, and whenever you need to know WHERE something is. Generated from the brain's real structure and always current.

• glossary

  The brain's map of known entities — people, initiatives, clients/projects, and all aliases — so you can link correctly ([[Entity Name]]) and know whether something the user mentions already exists. Consult before creating any entity note or writing notes that reference people/clients/projects.

• list_notes

  Browse the brain by folder prefix (e.g. "Knowledge/") — returns paths, newest first.

• move_note

  Rename or move a note to a new path AND repoint the [[wikilinks]] that pointed at it, so nothing breaks. Both the old and new locations must be within your write scope; refuses to overwrite an existing note. Notes outside your write scope cannot be rewritten — any of those you can READ that still link to the old path are listed back to you rather than left silently dangling; notes outside your rea

• overview

  Orient yourself: partitions with note counts, the most recently changed notes, and your access scope. Call at session start or when unsure where things live.

• read_note

  Read one note by its path relative to the brain root. Small notes return whole. For a LARGE note (a long reference doc), reading it whole floods your context — pass { section: "<heading>" } to read one section, or { offset, limit } to page through characters. With neither, a large note returns its outline (headings + offsets) + opening so you can pick what to read. When you deliberately need the E

• recall

  Deterministic retrieval over the brain. Returns evidence-tagged excerpts, a linked-context block (related notes up to 3 hops — read_note the ones you need), and a create-safety hint. If `clarify` is present, ASK THE USER which option they mean before proceeding — do not guess. If found=false, relay "not in the brain" honestly — never invent an answer.

• recent
• related

  Typed graph neighbors of a note (zero-LLM knowledge graph: wikilinks + frontmatter edges).

• remember

  One-step episodic memory record, filed per the schema and stamped with your identity.

• resolve

  Resolve a name/alias to its canonical note ("Makoto" → the note that owns that alias). Call before creating any entity note, and when the user references something by a nickname.

• search

  Fast ranked search for browsing and disambiguation — returns candidate notes with snippets, no committed answer. Use recall when you need a checked answer; use search when you need to find or choose between notes.

• skills

  The reusable SKILLS this brain carries — a skill is a packaged how-to the AI follows end-to-end (draft a proposal, build a client POC, prep for a meeting, etc.). Because they live in the vault, Callosium serves them to ANY connected AI: a skill authored in one assistant (Claude) works in another (ChatGPT/Gemini). Call this to see what is available, then read_note the skill's SKILL.md to follow it.

• write_notewrite action

  Create or update a note. Filing rules route new typed notes; attribution is stamped server-side. Writing to a path that ALREADY EXISTS is refused unless overwrite:true — use append_note to add safely, or pass overwrite:true to deliberately replace the body. For new entity notes, entity resolution runs first — if the entity already exists you get its canonical path back instead of a duplicate.

• Code scan34 source files scanned25/25
• –Live reliabilityno gateway calls yet and no remote to proben/a
• –Tool poisoningtools not inspected (local package is not executed); not countedn/a
• Auth qualitylocal package, no credentials required12/15
• Maintenancelast push 25 days ago15/15
• Maintainer identityregistry namespace matches repository owner7/10

Safety score

91/100

higher than 85% of 48,255 servers

Tools

21

the typical server has 7

Tools that change things

5%

typical: 0%. Fewer write tools means less an agent can break

GitHub stars

9

more than 87% of 28,688 servers

Downloads a week

31

more than 19% of 12,191 servers with a package

Callosium: common questions

Is Callosium MCP server safe?

Yes, by our scan: it is graded A (91/100). Read the Callosium safety report

How do I install Callosium?

It runs on your machine. Copy the Claude Code, Cursor, VS Code or Claude Desktop config from the install section.

Does Callosium need an API key?

No secret keys are declared. It reads 1 setting from the environment.

Is Callosium maintained?

The last commit was 66 days ago (2026-08-30). The latest release is v0.1.18.

How does Callosium compare with other MCP servers?

safety score 91/100 (higher than 85% of 48,255 servers); tools 21 (the typical server has 7); tools that change things 5% (typical: 0%. Fewer write tools means less an agent can break). Full comparison.