Documentation

Quickstart. No yak shaving.

ContextWell is a workspace + an MCP server. If you can sign in to a web app and paste a URL into your AI tool's connector settings, you can be up in five minutes.

Full reference docs are in progress. For anything not covered here, email [email protected].

Create your workspace

Sign up, pick a workspace name, and create your first knowledge base. Folder structure is up to you — most teams start with a /clients, /skills, /meetings split, but anything works.

Connect your AI tool via MCP

ContextWell exposes your workspace through a single MCP server. Plug it into Claude, ChatGPT, Cursor, Codex, Claude Code, or any custom agent — same URL, same tools, same workspace.

// In your harness's MCP / connector settings:
{
  "name": "contextwell",
  "transport": "http",
  "url": "https://contextwell.co/mcp",
  "auth": {
    "type": "bearer",
    "token": "YOUR_PERSONAL_ACCESS_TOKEN"
  }
}

Invite your team

Team plan includes unlimited collaborators — invite the engineers, designers, marketers, and ops folks who actually do the work. Everyone edits the same markdown surface in real time.

Let agents read and write

Once connected, your harness can list folders, search the KB, read full docs, and edit them — with comments and history preserved. Same files your humans see, same files your agents touch.

Try it

Prompts to start with.

Once your tool is connected, type these in a normal chat. Each one drives the MCP tools below — the same way in Claude, Cursor, ChatGPT, or Codex.

“Search my knowledge base for our onboarding flow and summarize the latest version.”

Finds the right doc and reads it back.

search_kb read_doc

“Read the notes in /meetings/q2-kickoff and draft a project brief in the Acme client folder.”

Turns raw notes into a new doc your team can edit live.

read_doc create_doc

“What did we decide about pricing, and which docs depend on that decision?”

Reasons over the knowledge graph — decisions and their backlinks.

search_kb graph_related

“Run the review-a-pr skill on this pull request.”

Invokes a team playbook as a slash command, the same way in every harness.

list_skills load_skill

MCP tools

What your AI tool can actually do.

Once connected, your harness has access to these tools. Same set across Claude, Cursor, ChatGPT, Codex — anything that speaks MCP.

list_kbs

List all knowledge bases in your workspace.

list_folder

List subfolders/docs as a tree, or pass `flat:true` for a flat sorted doc list.

search_kb

Full-text search across one or every enabled KB; each hit includes 1-hop graph neighbors.

read_doc

Read a doc's markdown body (optionally a line window), version, and metadata.

create_doc

Create a new doc in any folder.

edit_doc

Apply a single SEARCH/REPLACE edit to a doc.

multi_edit_doc

Apply multiple SEARCH/REPLACE edits in one call.

move_doc

Move or rename a doc.

graph_related

Walk the knowledge graph from a doc — set `direction:'in'` for backlinks.

graph_overview

Get a high-level graph view of a KB.

list_skills

List skill-kind playbooks the user has available.

load_skill

Load a skill body to follow as instructions.