RPCS-1

mcp

Connect the public RPCS-1 battery.

RPCS-1 provides a public Streamable HTTP MCP server with seven tools across three families: agent tuning, translation, and the per-user Translation Bridge loop. It is anonymous, deterministic, read-only, and does not require an API key.

Connection details

Name: RPCS1 Agent Tuner

URL: https://rpcs1.dev/mcp

Transport: Streamable HTTP

Authentication: None

Seven tools, three families

Agent tuning

recommend_agent_configuration — a five-primitive profile, failure-risk score, runtime posture, and next test for a described workload.

Translation

interpret (ambiguity detection and intent recovery), normalize (fragmented text to coherent prose), rewrite (six fixed audience styles).

Translation Bridge — the per-user loop

calibrate_profilebuilds a user's ReceiverProfile from five in-chat questions (continuous coordinates, never a category label). prepare_promptrecovers intent from the user's raw message on the way in. render_replyreturns deterministic rendering instructions for the user's profile on the way out. The profile travels as a parameter — nothing is stored server-side. Schema: /v1/receiver-profile.json

See examplesTry the tuner

First call to try

Use recommend_agent_configuration to diagnose my support copilot.

Task: refund and billing dispute triage
Environment: dynamic, somewhat_predictable, high stakes
Context relevance: medium
Commitment style: cautious
Target platform: anthropic

The useful first answer is the five-primitive profile, failure-risk score, runtime posture, and the next test to run.

The bridge loop, end to end

One calibration, then every reply adapts to you — no copy/paste. Paste this into any MCP client connected to the server:

Step 1 — once:
"Use calibrate_profile to build my receiver profile.
 Ask me the five questions, then save the profile JSON
 to your memory for this project."

Step 2 — every turn (add to project/system instructions):
"Before answering me, call render_reply with your draft
 and my saved profile, and apply its instructions.
 If my request is ambiguous, call prepare_prompt with my
 message and profile before acting."

Example: a profile of TI 20 · SG 25 · FT 80 · UE 75 · AR 75 makes every reply lead with the answer, stay flat and literal, welcome pushback, and commit to the most likely reading instead of interrogating you. The instructions are deterministic — same profile, same directives, every time.

Hyperagent compatibility

RPCS-1 provides a fixed public OAuth client for Hyperagent using authorization code flow with PKCE. In Hyperagent, enable Bring my own OAuth app and enter the discovery values below.

  • Redirect URI: https://hyperagent.com/api/mcp-servers/callback
  • Client ID: hyperagent-rpcs1
  • Client secret: leave blank
  • Authorization endpoint: https://rpcs1.dev/oauth/authorize
  • Token endpoint: https://rpcs1.dev/oauth/token
  • Scopes: mcp:tools

Safety and privacy

The MCP tool returns recommendations without modifying external systems. Only send task and environment information you are comfortable submitting to the RPCS-1 service.