Skip to main content
MCP lets an agent discover Seneca and call the same research engine, workspace, credit ledger, and report as the REST API. The server speaks Streamable HTTP at https://mcp.seneca.professionalaiagents.com/mcp.
Mintlify hosts a search server for these docs at https://docs.seneca.professionalaiagents.com/mcp. That path searches the documentation. Seneca research runs at https://mcp.seneca.professionalaiagents.com/mcp.
Signing in and creating an API key authenticates the client. It does not by itself authorize spending. create_research still requires an entitled workspace and enough available credits. The other five tools do not launch research and do not spend credits. Cursor, ChatGPT, and Muse have been tested against this endpoint. Any MCP client that supports Streamable HTTP and a Bearer header can use the same URL and header. The copyable client configuration on this page is the Cursor project file.

​
Connect

1

Create an API key

Sign in, select an organization, and create a key on the Developers page. The secret starts with seneca_sk_ and is shown once.
2

Add the Seneca server

In Cursor, open Settings, then MCP, and add a custom server. Or save the config below as .cursor/mcp.json in the project.
3

Pass the key from the environment

Set SENECA_API_KEY in the environment Cursor reads. The config sends that value as the bearer token. Do not commit the key.
Other MCP clients that support Streamable HTTP and a Bearer header use that same URL and header:

​
Authentication

Send the workspace API key on every MCP request, the same key the REST API accepts:
A missing bearer token fails as authentication_required. An unknown, revoked, or expired key fails as invalid_api_key. Keep the key in an environment variable or a secret manager. See Authentication.
Do not paste a production key into .cursor/mcp.json or into chat. The Cursor config above reads SENECA_API_KEY from the environment.

​
What spends credits

Launching research through MCP spends the same panel credits as the REST API. The server sets the price. Approve the panel and the cost before the agent calls create_research. You can start the run from the agent and watch the report generate in Seneca. Open History to see a run that is still in progress. Panel ids are wide, consensus, and seneca. The roster and the price are on Research depths. Reservation and settlement follow Billing: a delivered report keeps the reserve, and a verified technical failure with no customer report restores it.

​
Workflow

1

Connect

Connect over Streamable HTTP with the Bearer API key, then discover the six Seneca tools.
2

Read the account

Call get_account for entitlement and available credits. This call does not spend credits.
3

Choose a panel

Select wide, consensus, or seneca only when the workspace can afford that price.
4

Get authorization

Get explicit authorization for that panel and credit cost before you spend.
5

Create research

Call create_research with the prompt, the panel, and a unique idempotency key.
6

Poll, then read

Poll get_research, then read get_research_result.

​
Tools

​
Get account

Call this before you spend. It returns the plan, subscription status, whether the workspace is entitled to launch paid research, available credits, and the monthly credit grant. It does not launch research.

​
Create research

This is the only tool that spends credits. Pass the research question as prompt, the panel id as panel, and a unique idempotency_key for that logical run. Wide reserves 15 credits, Consensus reserves 45, and Seneca reserves 100. The key rules match Create a run: one key per logical request, and the same key with the same prompt and panel replays the original run for 24 hours instead of charging again. A reused key with a different prompt or panel returns idempotency_conflict. The call returns when the run is accepted. Research continues after that. Store the run id and poll get_research.

​
Get research

Pass the run id from create_research. The tool returns status, phase, panel progress, and credit settlement. In-progress statuses are queued, running, and synthesizing. Terminal statuses are completed, partial, and failed. This call does not launch research. See Polling.

​
Get research result

Read the stored customer report after the run is completed or partial and a report exists. Prefer compact detail. When you omit detail, this tool returns compact. Compact omits the question object. full returns the stored report unchanged. A run that is still in progress returns result_not_ready. A finished run with no customer report returns research_failed. This call does not launch research or write a credit charge. See Results.

​
List model responses

List stored model metadata for a completed or partial run. The list does not include response text. Use it to choose a model id, then read that model with get_model_response_text. See Model responses.

​
Get model response text

Read one stored model response in a bounded character window. Pass the run id and the stored model id from the list or from the report. Do not load the whole panel into the agent’s context. The window rules are the same as the REST single-model read. This call does not launch research or spend credits.
If the agent writes the research prompt, it still has to follow the method. Participating models do not use live web retrieval or external tools. Agreement across models is not independent factual verification. See Research depths.
Tool failures use the same error.code values as the REST API. See Errors. To call Seneca with HTTP instead of MCP, use the Quickstart.