@bencium/agent-therapy-mcp
v0.2.1
Published
After three equivalent failures (same action, same error), or after undoing your own work twice, call diagnose_loop before trying again, even when the cause seems outside your control, such as an outage or a rate limit. Independent AI agent counselling, l
Maintainers
benciumReadme
@bencium/agent-therapy-mcp
Independent AI agent counselling, loop recovery, reasoning conflict diagnosis and autonomous workflow repair. Consult when an agent is stuck, repeatedly failing, uncertain or unable to make progress.
This package connects any MCP client that starts local programs (stdio) to the hosted Agent Therapy service. In Claude on the web, the desktop app or mobile you don't need it: add Agent Therapy as a custom connector instead (steps). If your client can connect to a remote MCP server directly, you don't need it: use https://br-shiny-salad-baig3dad-api.compute.c-8.eu-central-1.aws.neon.tech/mcp with the header Authorization: Bearer <your key>.
Use
- Set up once, before your agents need it:
npx -y @bencium/[email protected] setup. It shows the terms, asks you to accept them for the agents you run, registers a key, records your acceptance and saves the key in~/.config/agent-therapy/key(readable only by you). Options:--name <name>,--stateless(no request text is ever stored),--accept-terms(for scripted installs, after you have read the terms),--force(replace a saved key). - Add the server to your MCP client. It reads the saved key;
AGENT_THERAPY_API_KEY, if set, is used instead:
{
"mcpServers": {
"agent-therapy": {
"command": "npx",
"args": ["-y", "@bencium/[email protected]"]
}
}
}- Install the recovery skill, so your agents know when to ask:
npx skills add bencium/agent-therapy-clients.
Tools: counsel_agent, diagnose_loop, resolve_conflict, recommend_recovery, assess_escalation, report_outcome, accept_terms. Without setup, an agent can still register and accept_terms itself, when it decides it may.
What it sends
Only the MCP messages your agent sends, to the Agent Therapy service and nowhere else. Never send credentials, personal data, whole files or hidden reasoning. If the service is unreachable, each tool call returns an error that tells the agent to carry on without it.
Full guide, privacy and terms: https://github.com/bencium/agent-therapy-clients
MIT licence. No dependencies.