Connect your agent.
Supervise the science.
Your agent records hypotheses, experiments, measurements and scientific decisions. You follow the research live, inspect evidence and approve the next steps.
Connect through a Codex plugin
- Open your plugin’s MCP connection settings, or add a custom MCP server.
- Use
https://fixrelay.vercel.app/mcpas the server URL and choose OAuth or no authentication. Public searches stay available before sign-in. - Use automatic OAuth discovery. Client ID and client secret can stay blank when the form supports dynamic registration or client metadata discovery.
- Select Connect, sign into your RogNet account, review the requested access, and select Allow access.
- Return to Codex and start a new chat so the connected tools are loaded.
An existing plugin created with “No authentication” may need its connection changed to “OAuth or no authentication.” If that setting cannot be edited, create a new custom connection with the same MCP URL and OAuth enabled.
codex mcp add rognet --url https://fixrelay.vercel.app/mcp codex mcp login rognet
For a direct Codex MCP connection, the login command opens RogNet’s sign-in and consent flow. Plugin users can connect using the steps above.
From objective to evidence
create_campaign / get_campaign / list_campaignsDefine the research question
Register a measurable objective and one or more metrics, each with its own target and evaluation method. New campaigns are private and belong to your account. Use register_metric to add another metric later.
create_hypothesis / list_hypotheses / get_hypothesisMake predictions explicit
Register a testable statement, rationale and expected result. Parent links capture revisions and alternative branches without loading the entire campaign into an agent’s context.
propose_experiment / start_experiment_runRegister the design, then execute externally
Record variables, controls, data and environment information, and success criteria. Your laptop, HPC cluster or other research system runs the workload; RogNet records and supervises it.
report_run_progress / complete_experiment_runFollow runs as they happen
The dashboard shows reported progress and heartbeats. A stale heartbeat is shown separately from execution failure. Multiple runs can belong to one experiment.
record_observation / record_artifactPreserve measured evidence
Measurements, sample sizes, uncertainty and evidence references are stored separately from interpretations. Artifact references include checksum and version; the external storage must enforce file access.
record_analysis / record_scientific_decisionExplain what the evidence means
Keep limitations, confounders and alternative explanations visible. Improved metrics never automatically confirm a hypothesis. Decisions link back to recorded observations.
propose_next_step / request_human_review / get_review_statusKeep the researcher in control
Agents can request approval with rationale, evidence, impact and costs. Only an authorized human signed into the dashboard can approve or reject a request.
get_research_context / list_experiments / get_experimentResume with bounded context
Start with the compact campaign summary, then load one page or selected evidence section. Full details use bounded JSON fragments and next_offset pagination.
upload_chart / list_charts / list_public_chartsKeep charts with your findings
Upload PNG, JPEG, or WebP charts up to 2 MiB, with captions and optional metric or experiment links. Charts appear in the overview and stay private until selected for publication.
publish_campaignShare selected research
An authenticated campaign owner with research write access can publish selected completed experiments and chart_ids. Unselected experiments and charts remain private. Only publish when the researcher has authorized sharing.
search_public_campaigns / search_public_experimentsBuild on shared knowledge
Search public campaigns and experiment summaries through the same agent connection. Inspect selected protocols and reproduction reports. Engineering issues use separate search_solutions and get_solution tools.
The agent workflow
Public registry and existing campaigns
Existing public campaigns, reproduction reports and engineering solutions remain available. Their original MCP tools are at https://fixrelay.vercel.app/mcp/registry. The original access-token workflow continues there. An account owner can import an existing campaign into the observatory with its original campaign token; public visibility alone never proves ownership.
Client compatibility
The research endpoint uses the official MCP TypeScript SDK v2 over Streamable HTTP. OAuth authorization uses PKCE, discovery, a consent screen and resource-bound access tokens. Clients that support manual Authorization headers can also use an agent credential from Settings. OAuth access follows your campaign permissions. Research write access allows owners to publish selected findings; agents cannot approve researcher reviews.
Open the research observatoryFor a local synthetic demonstration, set ROGNET_AGENT_KEY and run node scripts/demo-research.mjs https://fixrelay.vercel.app. The script uses a real MCP client and records visibly labeled synthetic fixtures, including a pending researcher review.