PJT AIPJT AI/MCP
Getting startedAPI
MCP integration
What is MCP?Connect Claude CodeConnect CursorConnect CodexContext scopeSecurity · permissions
MCP integration guide

Stream your context into AI dev tools

Reference PJT AI's project context directly from Claude Code · Cursor · Codex.

01What is MCP?#

MCP (Model Context Protocol) is a standard that lets AI models safely reference data in external systems. PJT AI runs a remote HTTP MCP server itself — with no install and no login, just register the URL and a token and Claude or Cursor reads your workspace data directly. It is read-only.

Query data from Claude Code with these tools:

ToolDescription
list_workspacesMy workspace list (start here when you don't know the ID)
get_projectProject basics (id or publicId p_xxx)
list_project_tasksTask list (status filter available)
get_taskTask details
get_documentDocument details (includes content)
list_workspace_documentsWorkspace document list
get_workspace_snapshotWorkspace bundle (recent documents, etc.)
ⓘ
Separate from running work
MCP is read-only. It works independently of task execution (pjt-agent start). Authentication reuses your existing Claude Code agent token (pjta_…) as an Authorization: Bearer header.

02Connect Claude Code#

Requirement: only the Anthropic claude CLI. No PJT npm install is needed.

1. Issue a token

Go to Settings → Claude Code agents → “Add agent” in your workspace and issue a token (pjta_…). Right after issuing, the screen shows registration snippets per tool with the token filled in (the token is shown only once).

2. Register — one line

Replace pjta_xxx with your token and paste it into the terminal.

bash
claude mcp add --transport http pjt-workspace https://mcp.pjt.ai \
  --header "Authorization: Bearer pjta_xxx"

3. Verify

bash
$ claude mcp list
pjt-workspace  ✓ connected

$ claude
> show me the recent documents in this workspace

[pjt-workspace] get_workspace_snapshot → 8 documents
✓
Connection successful
If claude mcp list shows pjt-workspace and you see the tool-call log in a conversation, you're connected.

03Connect Cursor#

Add the following to ~/.cursor/mcp.json. It uses the same token as Claude Code.

json
{
  "mcpServers": {
    "pjt-workspace": {
      "url": "https://mcp.pjt.ai",
      "headers": {
        "Authorization": "Bearer pjta_xxx"
      }
    }
  }
}
ⓘ
Restart required
Save and restart Cursor — the pjt-workspace tools then appear in the list.

04Connect Codex#

Codex does not support remote MCP natively yet, so mcp-remote bridges it. Add this to ~/.codex/config.toml.

toml
[mcp_servers.pjt-workspace]
command = "npx"
args = ["-y", "mcp-remote", "https://mcp.pjt.ai", "--header", "Authorization: Bearer pjta_xxx"]
ⓘ
Node.js required
The bridge runs through npx, so you need Node.js 18+ locally. Once Codex supports remote MCP, this becomes as simple as the Cursor setup.

05Context scope#

By default, MCP exposes only the context of projects the user is a member of. Finer-grained control is available as follows:

ScopeDefaultHow to control
Per projectMembership-basedWorkspace settings > Permissions
Sensitive fieldsIncludes budgets & quotesCan be excluded in settings
Meeting notesFull text exposedOption to expose summary only

06Security · permissions#

  • Read-only — MCP cannot change or delete data (query only)
  • Reuses the Claude Code agent token (pjta_) — no separate key; revoke/reissue per agent
  • The token is shown only once — reissue if lost. With the remote setup it lives only in your client config file; nothing from PJT is installed locally
  • Encrypted in transit and at rest (TLS 1.3 + AES-256); all access is recorded in audit logs
  • Not used for AI training — data ownership stays with the workspace
  • Enterprise can choose SSO/SAML and data residency (US · EU · KR)