Carbon Forge

Carbon Forge keeps code, files and prose in one record on the server, and people and agents read and write it through the same tools over MCP. Agents open lanes instead of cloning, every landing is one all-or-nothing changeset, and git is a projection pushed to GitHub. Public workspaces are free.

View on carbonforge.dev

Hosted MCP Server

npx add-mcp 'https://carbonforge.dev/api/mcp'

Installs into Claude Code, Codex, Cursor and more

Documentation

Your agent asks: who last changed this line, and why?

A repository

files · commits

$ git clone <repo>
Receiving objects: 
$ git blame -L ‹n›,‹n› src/index.ts
$ git log -1 ‹commit›
$ # why? the docs live in another tool

~100 GB history, 25M checkouts a day: Anthropic

Carbon Forge

code · pages · intent · lanes · seals · git

› trees.blame { "tree": "app", "path": "src/index.ts", "offset": ‹n›, "limit": 1 }
{ "path": "src/index.ts",
  "lines": [{ "line": ‹n›, "text": ‹text›, "revision_id": ‹id› }],
  "revisions": [{ "revision_id": ‹id›, "changeset_id": ‹id›,
    "intent": ‹intent›, "actor": ‹actor›, "edited_at": ‹time›,
    "lane": ‹lane› }], … }

One call. No clone. trees.blame

Not a repository. The record.

Code, knowledge and every change’s reason, sealed. Agents read and write it with a call; git is rendered from it.

166

MCP tools

clones to answer a question

changeset per landing, all or nothing

for public workspaces, forever

Read the docs

Proof, not promises.

Sealed changeset → git commit docs

tree ‹tree id›
parent ‹parent commit id›
author ‹actor› <‹actor›@users.noreply.‹domain›> ‹seconds› +0000
committer Carbon Forge <archive@users.noreply.‹domain›> ‹seconds› +0000

‹intent›

Carbon-Changeset: ‹changeset id›
Carbon-Tree-Revision: ‹tree revision id›
Carbon-Seal: ‹seal hash›
Carbon-Manifest: ‹manifest digest›

Stored once docs

› blobs.put { "content_base64": ‹bytes› }
{ "sha256": ‹digest›, "byte_size": ‹n›, "deduped": true }

Neighbours, built in docs

› kg.get_related { "pathOrId": "features/source-control-on-the-substrate", "kind": "page", "limit": 4 }
[ { "title": "Overview", "path": "overview", "score": 0.8935242, … },
  { "title": "Welcome to Carbon", "path": "home", "score": 0.8923878, … },
  { "title": "Source control: how it is built and why", "path": "architecture/source-control", "score": 0.8904409, … },
  { "title": "The Carbon paradigm", "path": "architecture/carbon-paradigm", "score": 0.86933243, … } ]

Pushed, deployed, read back docs

› archives.deploy_status { "tree": "app", "cut": "v1.4.0" }
{ "cut": "v1.4.0", "commit_id": ‹commit›, "summary": "success",
  "runs": [ { "name": ‹workflow›, "status": ‹status›, "conclusion": "success", "html_url": ‹url› } ] }

Why they leave Notion and Obsidian.

01 · XDA Developers

Plugins to maintain

Built into the record

:::atom{id=‹atom id›}   ← edit once, every page updates

02 · XDA Developers

Slower as it grows

Only the files you ask for

› trees.materialize { "tree": "app" }
{ "entries": [ { "path": ‹path›, "sha256": ‹digest›, "url": ‹signed url› } ] }

03 · XDA Developers

Search that cannot think

Meaning, not keywords

Related ▸ Overview · Welcome to Carbon · Source control: how it is built and why

04 · Notion Help

Locked in

Yours, byte for byte

.carbon/perspective.json
{ "format": "carbon-perspective/1",
  "entries": [ { "path": ‹path›, "revision_id": ‹id›, "seal_hash": ‹hash›, "sha256": ‹digest› } ] }

Bring your vault: Obsidian import