pebble-index-mcp
Enables voice captures from a smart ring to search and append notes in a local Obsidian vault and query an OpenAI-compatible AI agent, returning answers as phone notifications.
README
pebble-index-mcp
MCP bridge that exposes a local Obsidian vault and a Hermes Agent instance to the Pebble Index 01 smart ring's double-click MCP sandbox. Voice captures from the ring are routed by the Pebble app's cloud agent into this server's tools, and the answer comes back as a phone notification.
Architecture
[Index ring] → [Pebble app] → [Pebble cloud agent] → [this MCP server]
├─ vault tools (local files)
└─ ask_hermes → Hermes API server
- The server speaks MCP over Streamable HTTP (official
mcpSDK, FastMCP). - Every request must carry a bearer token; anything else gets 401.
- FastMCP's DNS-rebinding protection is on: allowed hosts are loopback plus
anything listed in
MCP_ALLOWED_HOSTS. - The public exposure is a cloudflared tunnel to the loopback port; no cloud infrastructure is required beyond that.
Components
| Path | Role |
|---|---|
src/pebble_index_mcp/vault.py |
Vault: path-sandboxed read/append + ripgrep search over the vault. All paths resolve inside the vault root; absolute paths, .. traversal, and symlink escapes are rejected. Appends are append-only, timestamped, and never overwrite. |
src/pebble_index_mcp/hermes.py |
HermesClient: forwards questions to an OpenAI-compatible chat completions endpoint with a short-answer system hint. Maps timeouts/transport failures/bad shapes onto HermesTimeout/HermesError. |
src/pebble_index_mcp/server.py |
FastMCP app: registers the four tools and the ring_persona prompt, wraps the streamable-http app in bearer auth (constant-time compare) and configures host allow-listing. |
Tools
vault_search(query, max_results=5)— case-insensitive text search; returnspath: excerptlines.vault_read(note_path, max_chars=1500)— reads the head of a note.vault_append(note_path, text)— appends a timestamped- HH:MM textline, creating the note if needed.ask_hermes(question)— forwards to the Hermes API server; 60s timeout with an honest acknowledgment on miss.
Security
- Auth: static bearer token (
MCP_BEARER_TOKEN), constant-time compared, required on every request. An empty token fails closed (all requests 401) and logs a warning at startup. - Transport: TLS terminates at the tunnel edge; the origin listens on loopback only.
- Host validation: DNS-rebinding protection rejects requests whose Host header is not loopback or in
MCP_ALLOWED_HOSTS. - Path sandbox: vault tools resolve all paths and require them to stay inside the vault root; symlink escapes are rejected and file opens use
O_NOFOLLOWon the final component. The sandbox guards against accidental and remote misuse; a hostile local process that races a parent-directory swap can still win (documented TOCTOU boundary — the vault is assumed to be a trusted single-user directory). - Config is read once at import: rotating the bearer token or changing allowed hosts requires a service restart.
- Secrets: the Hermes API key and MCP token live in a 0600 env file outside this repo, never in code, logs, or commits.
- Privacy note: recordings routed through the double-click sandbox transit the Pebble app's cloud agent. Single-click captures stay on-device. Sensitive thoughts belong on single-click.
Configuration (env)
| Var | Default | Purpose |
|---|---|---|
VAULT_PATH |
— | Required. Vault root for the file tools |
MCP_HOST / MCP_PORT |
127.0.0.1 / 8765 |
Listen address |
MCP_BEARER_TOKEN |
— | Required; request auth token |
MCP_ALLOWED_HOSTS |
— | Comma-separated extra allowed Host values (the public tunnel hostname goes here) |
HERMES_API_URL |
http://127.0.0.1:8642/v1/chat/completions |
ask_hermes target |
HERMES_API_KEY |
— | API server bearer key |
RING_MODEL |
pebble-ring |
Model alias sent upstream (map it to a cheap model via the API server's model_routes) |
RING_PERSONA_FILE |
— | Optional path to a text file replacing the generic cloud-agent persona |
HERMES_API_URL accepts any OpenAI-compatible chat completions endpoint, so
ask_hermes works against OpenRouter directly, Ollama, or any other
compatible API — Hermes is just the default.
Development
python3 -m venv .venv
.venv/bin/pip install -e ".[dev]"
.venv/bin/pytest -v
Run locally: set the env vars above and python -m pebble_index_mcp.server.
Deployment
deploy/ contains a systemd user unit template and a cloudflared tunnel
template (hostname and credentials filled in on the host, not in this repo).
See docs/ring-checklist.md for the phone-side setup and test matrix.
License
MIT — see LICENSE.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.