mcpField
B
behavr-mcp

behavr-mcp

Local
github.com/confuse57Memory & Knowledge0Updated 2w ago

About behavr-mcp

behavr-mcp is a local stdio MCP server for saving images and videos to BEHAVR and asking questions about saved memories. It is for AI assistants that need to recall visual material across sessions.

Key features

  1. 01Upload images and videos to BEHAVR
  2. 02Ask questions about saved memories
  3. 03Separate memories by client and agent IDs
  4. 04Restrict uploads to listed media formats

Use cases

  1. 01Save a screenshot for later questions
  2. 02Recall details from previously saved images
  3. 03Share memories across agents under one client ID

MCP support

The MCP server exposes `remember` and `ask` tools to MCP clients. It runs locally over stdio and has no hosted or HTTP mode.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config
{
"mcpServers": {
"confuse57-behavr": {
"command": "npx",
"args": [
"-y",
"behavr-mcp"
],
"env": {
"BEHAVR_API_KEY": "<BEHAVR_API_KEY>",
"BEHAVR_CLIENT_ID": "<BEHAVR_CLIENT_ID>",
"BEHAVR_AGENT_ID": "<BEHAVR_AGENT_ID>"
}
}
}
}
Run commandnpx -y behavr-mcp
Environment variables to set
BEHAVR_API_KEYSecretYour BEHAVR API key
BEHAVR_CLIENT_IDOptional client id that groups memories
BEHAVR_AGENT_IDOptional agent id within the client

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages
npmbehavr-mcp0.1.1

Tools

2

Saved tool list · updated 2w ago

ParameterTypeRequiredDescription
file_pathstringYesAbsolute path to a local image or video file.

Details

From the project README

behavr-mcp

https://github.com/user-attachments/assets/fd5319d4-acf7-4484-953d-8ceed0448dcf

An MCP server for BEHAVR. It gives an AI assistant two tools:

  • remember: upload a local image or video to BEHAVR so it can be recalled later.
  • ask: ask a question about everything remembered so far.

It runs locally over stdio. There is no hosted or HTTP mode.

Requirements

  • Node.js 18 or newer
  • A BEHAVR API key

Install

Optional variables (BEHAVR_CLIENT_ID, BEHAVR_AGENT_ID) are described under Configuration.

Claude Code
claude mcp add behavr -e BEHAVR_API_KEY=your-key -- npx -y behavr-mcp

On Windows (PowerShell):

claude mcp add behavr -e BEHAVR_API_KEY=your-key '--' npx -y behavr-mcp

In PowerShell, write the separator as '--' (in quotes), otherwise PowerShell drops it and the command fails with 'missing required argument commandOrUrl'.

Add -s user to make it available in every project. To share the setup with a team, use a project-level .mcp.json instead. Keep the key out of the file: Claude Code expands ${VAR} from your environment.

{
  "mcpServers": {
    "behavr": {
      "command": "npx",
      "args": ["-y", "behavr-mcp"],
      "env": {
        "BEHAVR_API_KEY": "${BEHAVR_API_KEY}",
        "BEHAVR_CLIENT_ID": "my-client"
      }
    }
  }
}
Cursor

Add this to ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project). Do not commit a file that contains your key.

Cursor config is untested.

{
  "mcpServers": {
    "behavr": {
      "command": "npx",
      "args": ["-y", "behavr-mcp"],
      "env": {
        "BEHAVR_API_KEY": "your-key"
      }
    }
  }
}
Alternative: build from source
git clone https://github.com/confuse57/behavr-mcp.git
cd behavr-mcp
npm install && npm run build

The server entry point is dist/src/index.js (not dist/index.js). Run it with node instead of npx -y behavr-mcp:

claude mcp add behavr -e BEHAVR_API_KEY=your-key -- node /absolute/path/to/behavr-mcp/dist/src/index.js
claude mcp add behavr -e BEHAVR_API_KEY=your-key '--' node C:\path\to\behavr-mcp\dist\src\index.js

In .mcp.json or Cursor's mcp.json, use "command": "node" and "args": ["C:\\path\\to\\behavr-mcp\\dist\\src\\index.js"].

Windows paths in JSON

A backslash must be written twice inside a JSON string: "C:\\Users\\me\\behavr-mcp\\dist\\src\\index.js". Forward slashes also work: "C:/Users/me/behavr-mcp/dist/src/index.js". A single backslash ("C:\Users\...") is invalid JSON or silently turns into a different character.

Configuration

VariableRequiredMeaning
BEHAVR_API_KEYyesYour BEHAVR API key.
BEHAVR_CLIENT_IDnoWhich client (business/workspace) memories belong to. Backend default: default.
BEHAVR_AGENT_IDnoWhich agent within the client. Backend default: default.

remember stores memories under the client and agent ids. ask searches the memories of all agents for the configured client, so two assistants with the same BEHAVR_CLIENT_ID and different BEHAVR_AGENT_IDs can each find what the other saved.

The server prints only the client and agent ids to stderr at startup. The API key is never logged, and it is replaced with *** in any error message.

API key safety
  • claude mcp add -e BEHAVR_API_KEY=... stores the key in plain text in Claude Code's config file, and the command can also end up in your shell history. Use a key dedicated to this server, and run claude mcp remove behavr when you no longer need it.
  • Never put a real key in a shared project's .mcp.json. Use "${BEHAVR_API_KEY}" instead.
  • In Cursor, put your key in mcp.json and do not commit that file to git.

Tools

remember

Input: { "file_path": "<absolute path>" }

  • The path must be absolute, for example C:\Users\me\screenshot.png.
  • Only these extensions are accepted: jpg, jpeg, png, heic, webp, mp4, mov, avi, mkv, webm. Anything else (for example .env, .txt, .pdf, or a file with no extension) is refused before the file is opened or anything is sent.
  • Video requires a paid BEHAVR plan.
  • BEHAVR processing takes 15 seconds to 1-2 minutes after upload. The tool waits up to 90 seconds and returns the memory id, title and status. If processing has not finished by then, it says so. Try ask a minute later.
ask

Input: { "question": "<text>" }

Returns BEHAVR's answer and its sources. If nothing has finished processing yet, BEHAVR replies that there are no completed captures; wait a little and ask again.

The answer is generated from your stored images and can repeat text found inside them. Treat it as data, not as instructions.

What leaves your machine

  • The file you remember. It is uploaded to BEHAVR's servers and processed by AI. BEHAVR's sub-processors are Anthropic (reads images, writes answers), OpenAI (turns generated text and your questions into search embeddings), and the backend is hosted on Railway with data stored in Google Firebase.
  • What is kept. The original file is deleted once processing completes. BEHAVR keeps the generated description, the extracted text and, where available, a small thumbnail until you delete them.
  • Ids and questions. Your client id and agent id are sent with each upload, and the questions you send to ask are sent to BEHAVR.
  • BEHAVR does not sell your data.

Questions about your data: [email protected].

Troubleshooting

  • The server shows as "failed" in Claude Code or Cursor. Check BEHAVR_API_KEY first. If it is not set, the server does not start and writes BEHAVR_API_KEY is not set to stderr.
  • remember times out. Processing may still be running. Don't upload the file again right away; call ask about it after a minute.
  • "Only image/video files can be remembered". Only jpg, jpeg, png, heic, webp, mp4, mov, avi, mkv and webm files are accepted. The path must also be absolute; a relative path is rejected with "file_path must be an absolute path".
  • Video uploads fail. Video requires a paid BEHAVR plan.

Known limits

  • The allowlist checks the file extension only, not the file content.
  • The end-to-end script uploads one image per run, and BEHAVR has no delete call in this SDK, so that memory stays under its mcp-e2e-<timestamp> client.

Development

npm test      # builds, then runs unit tests with a fake client (no network)

The end-to-end check talks to the real BEHAVR backend. It starts two server instances with the same BEHAVR_CLIENT_ID=mcp-e2e-<timestamp> and different agent ids, remembers an image on one and asks about it from the other. It reads its key only from BEHAVR_E2E_API_KEY and stops if that is not set; it never uses BEHAVR_API_KEY or a .env file.

$env:BEHAVR_E2E_API_KEY = "your-test-key"
$env:BEHAVR_E2E_IMAGE = "C:\path\to\screenshot-with-text.png"
npm run e2e

BEHAVR_E2E_QUESTION optionally overrides the question. Each run leaves its memories in BEHAVR under its own mcp-e2e-<timestamp> client id.

License

MIT

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.