@ai-dossier/mcp-server
Use one portable AI-agent skill across Claude Code, Codex, OpenCode, and MCP-compatible workflows.
@ai-dossier/mcp-server is for teams that want their MCP clients to discover, verify, and run trusted skills through the Model Context Protocol. Dossiers are portable, signed, versioned skills: familiar to Claude Code users, but verifiable and registry-backed for any compatible LLM tool.
Part of the ai-dossier ecosystem: use
@ai-dossier/clito author and install skills,@ai-dossier/corefor verification, and the ai-dossier project for the complete platform.
Installation
Claude Code Plugin (Recommended)
/plugin marketplace add imboard-ai/ai-dossier
/plugin install dossier-mcp-server@ai-dossier
One-time setup — auto-updates included.
Claude Code (Manual)
# Global (available across all projects)
claude mcp add dossier --scope user -- npx @ai-dossier/mcp-server
# Project-only
claude mcp add dossier -- npx @ai-dossier/mcp-server
Claude Desktop
Add to claude_desktop_config.json:
{
"mcpServers": {
"dossier": {
"command": "npx",
"args": ["-y", "@ai-dossier/mcp-server"]
}
}
}
Verify
claude mcp list
Then try:
"List available dossiers in this project"
"Execute the project-init dossier"
Tools
Discovery & Reading
| Tool | Description |
|---|---|
list_dossiers | Discover available dossiers in a directory. Scans for *.ds.md files and returns metadata. |
read_dossier | Read and parse a dossier file. Returns metadata and content. Should be called after verify_dossier passes. Pass dry_run: true to also get dry_run_plan (static preview). |
search_dossiers | Search the dossier registry for available dossiers by keyword and optional category filter. |
Security Verification
| Tool | Description |
|---|---|
verify_dossier | Verify integrity (checksum) and authenticity (signature) of a single dossier. Returns pass/fail with stage details. |
verify_graph | Batch verify all dossiers in a resolved dependency graph. Returns aggregate security report with per-dossier breakdown. |
Orchestration (Multi-Dossier Journeys)
| Tool | Description |
|---|---|
resolve_graph | Resolve a dossier dependency graph into an execution plan. Produces a DAG with ordered phases, parallel groups, and conflict detection. |
start_journey | Start a journey session from a resolved and verified graph. Returns the first step's dossier content with injected context. With dry_run: true no session is created and a static per-step preview (files, commands, network, env, risk score) is returned instead. The preview is derived from declared metadata and code blocks; the executing agent may take other actions and the score is a heuristic, not a safety guarantee. |
step_complete | Mark the current journey step as complete or failed. Advances to the next step with context from previous outputs. |
get_journey_status | Get the current state of a journey: completed steps, current step, remaining steps, and collected outputs. |
cancel_journey | Cancel an active journey session. Returns a summary of what completed before cancellation. |
Resources
| URI | Description |
|---|---|
dossier://concept | Introduction to dossiers — what they are and why to use them |
dossier://protocol | How to execute dossiers safely and effectively |
dossier://security | Security model, signing, verification, and trust management |
dossier://orchestration | Complete reference for multi-dossier journey tools |
Prompts
| Prompt | Description |
|---|---|
execute-dossier | Run a dossier with verification and protocol. Automatically chooses single or multi-dossier flow based on relationships. |
execute-journey | Guide through a multi-step dossier journey: resolve, verify, present plan, execute steps, collect outputs. |
create-dossier | Author a new dossier using the official meta-dossier template. |
Security
Every dossier execution is verified before running:
User: "Execute the deploy-to-aws dossier"
LLM: *Calls verify_dossier*
Security Verification:
- Integrity: Checksum verified (content not tampered with)
- Authenticity: Signed by imboard-ai-2024 (trusted)
- Risk Level: HIGH
This dossier will modify AWS infrastructure.
Proceed? (y/N)
Verification results
- ALLOW — Verified signature from trusted source + low risk. Execute confidently.
- WARN — Unsigned/unknown signer or high risk. Request user approval.
- BLOCK — Checksum failed or signature invalid. Do not execute.
Features
- Integrity: SHA256 checksums ensure dossiers haven't been modified
- Signatures: Optional Ed25519 or AWS KMS cryptographic signatures (legacy minisign keys still verify)
- Risk assessment: Declared risk level and specific risk factors per dossier
- Trust model: Decentralized — users choose which signing keys to trust via
~/.dossier/trusted-keys.txt
Development
git clone https://github.com/imboard-ai/ai-dossier.git
cd ai-dossier/mcp-server
npm install
npm run build
npm test
npm start
Project structure
mcp-server/
src/
index.ts # Server entry point
tools/ # Tool implementations
resources/ # Resource providers
orchestration/ # Graph resolution, sessions, output mapping
utils/ # Logger, CLI wrapper, response helpers
Technical Details
- Protocol: MCP (Model Context Protocol)
- Language: TypeScript
- Runtime: Node.js 20+
- Transport: stdio
- Dependencies:
@modelcontextprotocol/sdk,@ai-dossier/core - License: AGPL-3.0
For AI Agents
npm run build -w mcp-server # build (requires core built first)
npm run test -w mcp-server # test
make build-mcp # build core + mcp-server together
src/index.ts— server entry: registers all tools, resources, and promptssrc/tools/— one file per MCP tool (e.g.verifyDossier.ts)src/resources/— static MCP resources (concept, protocol, security, orchestration)src/orchestration/— graph resolution and journey session managementsrc/utils/— logger, CLI wrapper, response helpers
Adding a tool: create src/tools/myTool.ts, export handler + input type, register in src/index.ts (add to ListTools + CallTool switch).
Adding a resource: create src/resources/myResource.ts, register in src/index.ts (add to ListResources + ReadResource switch).
For dossier users (not contributors): install via claude mcp add dossier -- npx @ai-dossier/mcp-server to get dossier:// resources and tools at runtime. MCP Registry name: ai.imboard/dossier.