MCP
ingence mcp lets an agent ask Ingence questions about one project. It uses the Model Context
Protocol (MCP), a standard way for an agent to ask a program questions. Ingence answers from the
project’s last scan. It doesn’t change the project, and it doesn’t send the contents of your files.
Use Ingence with an AI agent says what an agent is and how to bring its
answers back.
Start it
Your agent starts the command itself. You don’t run it in a terminal. Add Ingence to your agent’s
list of MCP servers, with your project’s folder in place of <project>:
{
"mcpServers": {
"ingence": {
"command": "ingence",
"args": ["mcp", "-p", "<project>", "--level", "words"]
}
}
}
Each agent keeps this list in its own settings file. Its documentation says where. If your agent
can’t find ingence, give the full path to the program. which ingence prints it on macOS and
Linux, and where ingence prints it on Windows.
The command takes two options:
Option |
What it does |
|---|---|
|
The project’s folder. Without it, Ingence looks in the current folder and the folders above it. |
|
How much the answers name: |
The levels are the same as for the summary (Choose a level). At every level, Ingence leaves out your files’ contents and full paths. It also leaves out your notes and the names of the people who answered.
What the agent can ask
Each question is a tool. Ingence answers each one with one block of text in JSON.
Tool |
What Ingence answers |
Arguments |
Level |
|---|---|---|---|
|
The project on one screen: the number of experiments, files and areas, the dates, file types, completeness, repeats, the main themes, goals and how much the scan read. |
none |
every level |
|
The experiments, in date order, one page at a time. Each has its short name, dates, theme, files, completeness and repeats. |
|
every level |
|
One experiment, with its story day by day: its workflow, its repeats and the files it was made from. |
|
|
|
The same answer as Ask in the Dashboard: how Ingence read the question, the experiments that match it, best first, and the experiments in the same theme. |
|
every level |
|
Each experiment’s raw files and how many have an analysis, with totals for each person and each theme. |
none |
every level |
|
The themes: the words of each, how many experiments it holds and its years. |
none |
every level |
|
Your own topics and goals, and the experiments that match each. |
none |
every level |
|
The assumptions the results rest on, with how many files rest on each. |
none |
every level |
|
The checks that wait for your answer, with the answers each one takes. |
none |
every level |
|
One experiment’s files, one page at a time: each file’s name, its path inside its data folder, its size and type, and what Ingence read it as. |
|
|
offset is the number of rows to skip, 0 by default. limit is the most rows to give: 50 by
default, and at most 500.
Short names
Ingence gives each experiment a short name, such as E1, E2 and E3. It numbers the experiments
in date order, the first time the agent asks for them. The names last until the agent stops the
command. Areas, themes, goals, topics, assumptions and questions get short names that start with
A, T, G, K, S and Q. The project’s own identifiers stay in the project.
What Ingence answers
Ingence answers only the tools in the table. The tools don’t write to the project. Below the paths
level, experiment_files isn’t on the list, and Ingence refuses a request for it.
When Ingence refuses a request, it says why in words, such as no experiment is called E99; list_experiments gives the aliases.
The protocol
For a person who writes an MCP client:
Ingence reads JSON-RPC 2.0 messages on standard input, one message to a line. It writes one answer to a line on standard output. It doesn’t open a network connection.
It answers
initialize,ping,tools/listandtools/call. It accepts any notification and doesn’t answer it.It speaks the protocol versions
2025-06-18,2025-03-26and2024-11-05. When a client asks for another version, Ingence answers with2025-06-18.The command stops when its input closes.