MCP tools reference

CodeSage serves seven tools and two resources to any MCP client. Claude Desktop reaches them through a stdio bridge that forwards each call to the backend over HTTP. This page lists every argument, every result field and every error message the tools produce. For the HTTP side of the same calls, see MCP HTTP endpoints. For setup, see Claude Desktop.

Argument schemas live in backend/codesage_mcp/schemas.py. The bridge answers tools/list from that file, so the list works even while the backend is down. Only tool calls and resource reads need the backend.

Overview

Tool	Reads from	Needs a completed repository	Required arguments	What it returns
list_repositories	SQLite	No	none	Every repository with status and size
search_code	Algolia	Only if repository is given	query	Ranked hits with file locations, snippets and source links
get_file	SQLite	Yes	repository, path	A file with line numbers, at most 400 lines per call
list_definitions	SQLite	Yes	repository	Entities ordered by file and line
analyze_repository	SQLite	Yes	repository	Counts, languages, top directories, largest files
find_patterns	SQLite	Yes	repository, pattern	Lines matching a literal string or regular expression
add_repository	Git host, SQLite	No	url	The new repository’s ID and status; indexing continues in the background

Every tool call counts toward the rate limit of 30 per minute per client address. See Configuration.

Shared rules

• Unknown arguments are rejected. Every argument model sets additionalProperties: false. The bridge checks arguments first, and the backend checks them again.
• Failures are tool errors. A failure the caller can act on (an unknown repository, a repository still indexing) comes back as a tool result with is_error: true and a one-line message. Transport and HTTP failures also come back as tool errors; see How failures reach Claude.
• Results are JSON text. The bridge returns the result object as pretty-printed JSON in one text block.

Resolving the repository argument

Five tools take a repository argument. It accepts three forms, and the backend resolves it the same way for all of them.

You pass	Matches
12	The repository whose numeric ID is 12
octo/demo	A repository whose name, or whose GitHub owner/name taken from its URL, equals the text, ignoring case
gitlab.com/group/project	A repository named <host>/<path>, which is how repositories on other hosts are named. The stored URL (https://gitlab.com/group/project) also matches

The rules, in order:

1. The text is trimmed.
2. If it is made only of ASCII digits, it is looked up as an ID and never as a name. A repository whose name is a number can only be reached by its ID. An all-digit value longer than 19 digits is treated as not found, because no ID can be that long.
3. Otherwise the text is lowercased and compared with each repository’s name and URL path. One match wins.
4. Several matches are an error that lists the candidates. Use the numeric ID.
5. No match is an error that lists the repository names that exist.

After resolution, every tool except list_repositories and add_repository checks that the repository is ready.

Situation	Error message
No match	Repository 'nope' not found. Available repositories: gitlab.com/gitlab-org/gitlab-test, pmndrs/zustand
No match and nothing indexed	Repository 'nope' not found. Available repositories: none indexed yet
More than one match	Repository 'demo' is ambiguous: demo (id 3), demo (id 7). Use the numeric ID.
Status failed	Repository 'octo/demo' failed to index: <error_message>. Reprocess it from the dashboard. The : <error_message> part is left out when there is none
Status pending or processing	Repository 'octo/demo' is still indexing (120/400 files); try again when it completes. The (n/m files) part appears only when the total file count is known

Note

A repository that finished with Partially indexed: ... in error_message has status completed, so the tools accept it. Read error_message in list_repositories to see what was left out.

list_repositories

Lists every repository CodeSage knows about, ordered by ID. It does not require a ready repository, so it also shows ones that are indexing or failed. Use it to see IDs, names and progress.

Arguments

None. The schema is {"type": "object", "properties": {}, "additionalProperties": false}.

Result

Field	Type	Meaning
repositories	array	One object per repository
repositories[].id	integer	Numeric ID
repositories[].name	string	owner/name for GitHub, <host>/<path> for other hosts
repositories[].url	string or null	Normalised https:// URL
repositories[].branch	string or null	null means the default branch
repositories[].commit_sha	string or null	The commit the index was built from. Source links point at it. null until a run has listed the files, or for a repository indexed before commits were recorded
repositories[].status	string	pending, processing, completed or failed
repositories[].total_files	integer	Files selected for the last run
repositories[].total_entities	integer	Entities stored in SQLite
repositories[].error_message	string or null	Why the last run failed, or what a completed run left out
repositories[].processed_at	string or null	ISO 8601 time the last run finished, in UTC with a +00:00 offset

The real result, from a backend holding one finished and one failed repository. Both were indexed before commits were recorded, so commit_sha is null:

Result

{
  "repositories": [
    {
      "id": 1,
      "name": "gitlab.com/gitlab-org/gitlab-test",
      "url": "https://gitlab.com/gitlab-org/gitlab-test",
      "branch": null,
      "commit_sha": null,
      "status": "failed",
      "total_files": 0,
      "total_entities": 0,
      "error_message": "Processing was interrupted by a server restart; reprocess the repository",
      "processed_at": null
    },
    {
      "id": 2,
      "name": "pmndrs/zustand",
      "url": "https://github.com/pmndrs/zustand",
      "branch": null,
      "commit_sha": null,
      "status": "completed",
      "total_files": 120,
      "total_entities": 1025,
      "error_message": null,
      "processed_at": "2026-10-26T23:32:50.170474+00:00"
    }
  ]
}

Limits and errors

The list is not paged or truncated. This tool has no tool-level errors beyond the transport failures in How failures reach Claude.

Example prompt

Which repositories does CodeSage have indexed, and are they all finished?

search_code

Full-text search over the Algolia index of functions, classes (including interfaces, type aliases and enums), Markdown sections and Terraform blocks. Imports are searched only when entity_type is import. Without repository, it searches every indexed repository.

Arguments

Argument	Type	Required	Default	Constraints
query	string	Yes	none	minLength 1, and at most 512 bytes of UTF-8
repository	string or null	No	all repositories	ID, name or owner/name; the repository must be completed
language	string or null	No	none	Exact language label such as python, typescript, cpp, csharp. See Languages
entity_type	string or null	No	none	One of function, class, import, doc_section, terraform_block
limit	integer	No	10	1 to 50. Results per page
page	integer	No	0	0 or more. Zero-based

How the search runs

• Each query asks Algolia to drop words that miss when nothing matches every word (removeWordsIfNoResults: allOptional), ignore stop words and plurals, and use English (queryLanguages: ["en"]). A question in plain words returns the best partial matches instead of nothing.
• repository, language and entity_type become filters joined with AND. Values are quoted and escaped, so input cannot add filter clauses.
• Without entity_type, the filter NOT entity_type:"import" is added, so import lines don’t crowd out the definitions. Pass entity_type: "import" to search imports.
• Part of a camelCase name matches: WithState finds TestShallowWithState.
• On a relevance tie, source files rank before test files, then rank_weight decides the order: classes, then functions and Terraform blocks, then doc sections, then imports. See Algolia record fields.
• With no repository, hits can come from repositories that are being reprocessed.

Result

Field	Type	Meaning
query	string	The query as sent
repository	string or null	Name of the repository you filtered by, otherwise null
total_hits	integer	Total matches reported by Algolia (nbHits)
page	integer	The page returned
pages	integer	Total pages (nbPages)
results	array	Hits for this page
results[].entity_name	string	Bare name, for example parse_repository_url
results[].entity_type	string	One of the five entity types
results[].file_path	string	Path inside the repository
results[].line_start, line_end	integer	1-based line range
results[].language	string	Language label of the file
results[].signature	string or null	The definition line, cut at 300 characters; null when empty
results[].snippet	string	Signature plus body, cut at 600 characters
results[].description	string or null	For example Python function: GitSource._read_tree, cut at 300 characters; null when empty
results[].url	string or null	Source link at the exact lines, or null when the host has no known link format

Source link formats:

Host	Link
github.com	https://github.com/<owner>/<repo>/blob/<ref>/<path>#L<start>-L<end>
gitlab.com and hosts starting with gitlab.	https://<host>/<path>/-/blob/<ref>/<file>#L<start>-<end>
bitbucket.org	https://bitbucket.org/<path>/src/<ref>/<file>#lines-<start>:<end>
Any other host	null

<ref> is the commit the index was built from (commit_sha in list_repositories), so the line numbers stay right after the branch moves on. A repository indexed before commits were recorded uses its branch, or HEAD when it has none, until it is reprocessed. Path segments are percent-encoded.

Real output for query: "createStoreImpl", repository: "pmndrs/zustand" and limit: 2, from a repository indexed before commits were recorded (hence HEAD in the links). The snippet strings are trimmed here with …:

Result

{
  "query": "createStoreImpl",
  "repository": "pmndrs/zustand",
  "total_hits": 5,
  "page": 0,
  "pages": 3,
  "results": [
    {
      "entity_name": "createStoreImpl",
      "entity_type": "function",
      "file_path": "src/vanilla.ts",
      "line_start": 60,
      "line_end": 97,
      "language": "typescript",
      "signature": "const createStoreImpl: CreateStoreImpl = (createState) => {",
      "snippet": "const createStoreImpl: CreateStoreImpl = (createState) => {\n  type TState = ReturnType<typeof createState>\n  type Listener = (state: TState, prevState: TState) => void\n  let state: TState\n  const listeners: Set<Listener> = new Set…",
      "description": "TypeScript function: createStoreImpl",
      "url": "https://github.com/pmndrs/zustand/blob/HEAD/src/vanilla.ts#L60-L97"
    },
    {
      "entity_name": "setState",
      "entity_type": "function",
      "file_path": "src/vanilla.ts",
      "line_start": 66,
      "line_end": 81,
      "language": "typescript",
      "signature": "const setState: StoreApi<TState>['setState'] = (partial, replace) => {",
      "snippet": "  const setState: StoreApi<TState>['setState'] = (partial, replace) => {\n    // TODO: Remove type assertion once https://github.com/microsoft/TypeScript/issues/37663 is resolved\n    // https://github.com/microsoft/TypeScript/issue…",
      "description": "TypeScript function: createStoreImpl.setState",
      "url": "https://github.com/pmndrs/zustand/blob/HEAD/src/vanilla.ts#L66-L81"
    }
  ]
}

After a reprocess, each url has the indexed commit’s SHA where this one has HEAD.

Limits and truncation

• query is at most 512 bytes of UTF-8, because Algolia rejects longer queries.
• limit is capped at 50 per page. Use page for more.
• Snippets are cut at 600 characters, and signatures and descriptions at 300.
• The index stores at most 2,000 characters of each entity’s body, so a snippet never shows more than that.

Errors

Message	Cause
Any message from Resolving the repository argument	The repository is unknown, ambiguous, failed or still indexing
Invalid arguments for 'search_code': ... ending in query is too long; the limit is 512 bytes	query is over 512 bytes of UTF-8
Search is not configured on the server (ALGOLIA_APP_ID / ALGOLIA_ADMIN_API_KEY are missing).	The backend has no Algolia credentials
Search is unavailable; try again later.	Algolia rejected the query or could not be reached. The detail is logged on the backend only, because it can name the Algolia host or app ID

Example prompt

Search the octo/demo repository for where the Algolia index settings are applied.

get_file

Reads one stored file with line numbers. The text comes from SQLite, so it matches what was indexed, not the live branch.

Arguments

Argument	Type	Required	Default	Constraints
repository	string	Yes	none	ID, name or owner/name; must be completed
path	string	Yes	none	Exact path, case-sensitive, relative to the repository root
start_line	integer or null	No	1	1 or more
end_line	integer or null	No	end of file	1 or more

Result

Field	Type	Meaning
repository	string	Repository name
path	string	The stored path
language	string	Language label
total_lines	integer	Lines in the whole file
start_line	integer	First line returned
end_line	integer	Last line returned
truncated	boolean	true if the 400-line or 100,000-character cap cut the output short
content	string	Lines formatted as <number>| <text>, with the number right-aligned to at least 4 characters

Real output for repository: "pmndrs/zustand", path: "src/vanilla.ts", start_line: 1 and end_line: 6:

Result

{
  "repository": "pmndrs/zustand",
  "path": "src/vanilla.ts",
  "language": "typescript",
  "total_lines": 100,
  "start_line": 1,
  "end_line": 6,
  "truncated": false,
  "content": "   1| type SetStateInternal<T> = {\n   2|   _(\n   3|     partial: T | Partial<T> | { _(state: T): T | Partial<T> }['_'],\n   4|     replace?: false,\n   5|   ): void\n   6|   _(state: T | { _(state: T): T }['_'], replace: true): void"
}

Limits and truncation

• At most 400 lines per call (MAX_FILE_LINES). The returned end_line is the smaller of your end_line, the last line and start_line + 399.
• At most 100,000 characters of formatted content.
• When truncated is true, call again with start_line set to the returned end_line + 1.

Errors

Message	Cause
File 'src/mian.py' not found in octo/demo. Close paths: src/main.py, src/main_test.py	No stored file has that exact path. Up to 10 similar paths (similarity cutoff 0.4) are suggested. The Close paths: part is left out when none are close
start_line 500 is past the end of the file (104 lines)	start_line is greater than the line count
end_line must not be before start_line	end_line is smaller than start_line
Repository errors	See Resolving the repository argument

Example prompt

Show me lines 1 to 120 of backend/app/main.py in octo/demo.

list_definitions

Lists stored entities for one repository, ordered by file path, then start line, then ID. It reads SQLite, so it is exact and unranked.

Arguments

Argument	Type	Required	Default	Constraints
repository	string	Yes	none	ID, name or owner/name; must be completed
path_prefix	string or null	No	none	Only files whose path starts with this text. Literal match, no wildcards
entity_type	string or null	No	none	function, class, import, doc_section or terraform_block
name	string or null	No	none	Case-insensitive substring of the entity name. Literal match
limit	integer	No	100	1 to 500

Result

Field	Type	Meaning
repository	string	Repository name
total	integer	All entities that match the filters, not only those returned
definitions	array	Up to limit entities
definitions[].entity_type	string	One of the five entity types
definitions[].entity_name	string	Bare name; methods are not prefixed with their class
definitions[].file_path	string	Path inside the repository
definitions[].line_start, line_end	integer	1-based range
definitions[].signature	string or null	Cut at 300 characters
definitions[].description	string or null	The stored description, cut at 300 characters

Real output for repository: "pmndrs/zustand", path_prefix: "src/vanilla" and limit: 3. total is 14, so the response holds the first three of 14 matches:

Result

{
  "repository": "pmndrs/zustand",
  "total": 14,
  "definitions": [
    {
      "entity_type": "class",
      "entity_name": "StoreApi",
      "file_path": "src/vanilla.ts",
      "line_start": 9,
      "line_end": 14,
      "signature": "export interface StoreApi<T> {",
      "description": "TypeScript class: StoreApi"
    },
    {
      "entity_type": "class",
      "entity_name": "StoreMutators",
      "file_path": "src/vanilla.ts",
      "line_start": 40,
      "line_end": 40,
      "signature": "export interface StoreMutators<S, A> {}",
      "description": "TypeScript class: StoreMutators"
    },
    {
      "entity_type": "function",
      "entity_name": "createStoreImpl",
      "file_path": "src/vanilla.ts",
      "line_start": 60,
      "line_end": 97,
      "signature": "const createStoreImpl: CreateStoreImpl = (createState) => {",
      "description": "TypeScript function: createStoreImpl"
    }
  ]
}

This output comes from an index built before TypeScript type aliases and enums were extracted. After a reprocess, StoreApi and StoreMutators are described as TypeScript interface: ..., type aliases (TypeScript type: ...) and enums (TypeScript enum: ...) are listed as class entities too, and total grows accordingly.

Limits and errors

limit caps the page at 500 and there is no paging, so narrow the filters when total is larger than the number returned. Errors are the repository errors from Resolving the repository argument.

Example prompt

List every class in octo/demo under backend/app/services.

analyze_repository

Returns measured counts and layout, with no interpretation. Use it first to learn what a repository contains.

Arguments

Argument	Type	Required	Constraints
repository	string	Yes	ID, name or owner/name; must be completed

Result

Field	Type	Meaning
repository	string	Repository name
repository_id	integer	Numeric ID
status	string	Always completed here
total_files	integer	Files stored
processed_files	integer	Files processed in the last run
total_lines	integer	Lines across stored files
total_entities	integer	Entities stored
entities_by_type	object	Entity count per type, for example {"function": 910}
files_by_language	object	File count per language label
language	string or null	Primary programming language chosen by the last run
framework	string or null	Framework guessed from package.json or Python imports
top_directories	array	Up to 10 {directory, files} pairs by file count. The directory is the first path segment, or (root) for files at the top
largest_files	array	The 10 files with the most lines, as {file_path, line_count}
definitions_by_language	object	Entity count per file language. A file with an unknown language is grouped under unknown

Real output for repository: "pmndrs/zustand". The top_directories and largest_files lists are trimmed to three entries (the tool returns up to 10):

Result

{
  "repository": "pmndrs/zustand",
  "repository_id": 2,
  "status": "completed",
  "total_files": 120,
  "processed_files": 120,
  "total_lines": 25117,
  "total_entities": 1025,
  "entities_by_type": {
    "class": 22,
    "doc_section": 398,
    "function": 474,
    "import": 131
  },
  "files_by_language": {
    "css": 3,
    "html": 2,
    "javascript": 14,
    "json": 7,
    "markdown": 47,
    "typescript": 34,
    "yaml": 13
  },
  "language": "typescript",
  "framework": "react",
  "top_directories": [
    {
      "directory": "docs",
      "files": 42
    },
    {
      "directory": "examples",
      "files": 27
    },
    {
      "directory": "src",
      "files": 16
    }
  ],
  "largest_files": [
    {
      "file_path": "tests/devtools.test.tsx",
      "line_count": 2926
    },
    {
      "file_path": "docs/learn/guides/tutorial-tic-tac-toe.md",
      "line_count": 1429
    },
    {
      "file_path": "tests/persistAsync.test.tsx",
      "line_count": 1244
    }
  ],
  "definitions_by_language": {
    "javascript": 64,
    "markdown": 398,
    "typescript": 563
  }
}

Limits and errors

The lists are capped at 10 entries. Errors are the repository errors from Resolving the repository argument.

Example prompt

Analyze octo/demo: how big is it, which languages does it use and where is the code?

find_patterns

Finds lines in the stored files that match a literal string or a regular expression. Use it when search ranking is not what you want, for example to find every call to a function. Files are scanned in path order.

Arguments

Argument	Type	Required	Default	Constraints
repository	string	Yes	none	ID, name or owner/name; must be completed
pattern	string	Yes	none	1 to 200 characters
regex	boolean	No	false	true treats pattern as a regular expression
path_glob	string or null	No	none	Matched against the whole path, case-sensitive
max_results	integer	No	100	1 to 500

Matching rules:

• A literal pattern is a case-sensitive substring test on each line.
• A regular expression is compiled with the Python regex package and searched in each line. It is not anchored, so use ^ and $ if you need that. Add (?i) for a case-insensitive match.
• path_glob uses fnmatch semantics on the full path, and * also matches /. src/*.py therefore matches src/a/b.py.

Result

Field	Type	Meaning
repository	string	Repository name
pattern	string	The pattern as sent
regex	boolean	Whether it was treated as a regular expression
total_matches	integer	Number of matches returned (never more than max_results)
truncated	boolean	true if more matches exist, or if a regex time budget ran out
matches	array	{file_path, line, text} objects. text is the line with surrounding whitespace removed, cut at 300 characters
skipped_files	array	Present only when files were skipped. The first 50 paths
skipped_count	integer	Present only when files were skipped. The full count

Real output for repository: "pmndrs/zustand", pattern: "createStoreImpl" and max_results: 3:

Result

{
  "repository": "pmndrs/zustand",
  "pattern": "createStoreImpl",
  "regex": false,
  "total_matches": 2,
  "truncated": false,
  "matches": [
    {
      "file_path": "src/vanilla.ts",
      "line": 60,
      "text": "const createStoreImpl: CreateStoreImpl = (createState) => {"
    },
    {
      "file_path": "src/vanilla.ts",
      "line": 100,
      "text": "createState ? createStoreImpl(createState) : createStoreImpl) as CreateStore"
    }
  ]
}

Limits and truncation

Limit	Value	Effect
Matches	max_results (1 to 500)	The scan stops at the first extra match and sets truncated: true
Regex time per file	1 second	The file is added to skipped_files; the scan goes on
Regex time per call	10 seconds	The remaining files are added to skipped_files and truncated is true
Regex file size	Files over 1,000,000 characters	Skipped and listed in skipped_files
Literal search	No time limit	Not subject to the regex limits

Errors

Message	Cause
Invalid regular expression: <detail>	regex is true and pattern does not compile
Repository errors	See Resolving the repository argument

Example prompt

In octo/demo, find every place that matches TODO|FIXME in Python files under backend/, as a regular expression.

add_repository

Adds a repository by https URL and starts indexing it in the background. It works with GitHub, GitLab (nested groups too), Bitbucket and self-hosted git servers. It runs the same pre-flight checks as the REST endpoint, and nothing is created if a check fails.

Arguments

Argument	Type	Required	Default	Constraints
url	string	Yes	none	minLength 1. An https repository URL. http:// and git@host:path forms are rewritten to https://. A branch URL such as .../tree/dev, .../-/tree/dev or Bitbucket’s .../src/dev is understood
branch	string or null	No	default branch	At most 100 characters. Must not contain a .. path segment or a NUL, and must not start with -. An explicit value wins over a branch in the URL
description	string or null	No	none	Free text note

The URL is refused if it has a port, credentials, a query or fragment, the host localhost (or *.localhost), or a private or loopback IP address. A path that is not owner/repo (plus a recognised branch suffix) is refused as well. On hosts other than github.com, the host name is also resolved before git runs: if it doesn’t resolve, or any address it resolves to is not public, the repository is refused, and git then connects only to the address that was checked.

Private repositories work only on github.com, and only when the server’s GITHUB_TOKEN can read them. On every other host the repository must be public.

Result

Field	Type	Meaning
repository_id	integer	ID to pass to the other tools
name	string	owner/name for GitHub, <host>/<path> otherwise
status	string	pending at creation
message	string	Always Indexing started; check list_repositories for progress.

The values below are illustrative. This call writes, so it was not run against a live backend; the shape follows the response schema in the code.

Result

{
  "repository_id": 5,
  "name": "gitlab.com/group/project",
  "status": "pending",
  "message": "Indexing started; check list_repositories for progress."
}

Indexing runs after the tool returns. Poll list_repositories until the status is completed, then search it.

Limits

The pre-flight check takes up to 10 seconds. A later clone on a non-GitHub host takes up to 120 seconds. Indexing keeps at most 1,000 files and 100 MB, and skips files of 1 MB or more. See Indexing pipeline.

Errors

Message	Cause
url: <reason>	The URL failed validation, for example Invalid repository URL (only https is supported): ...
branch: Invalid branch name	The branch contains .., a NUL, or starts with -
octo/demo is already added as repository 1.	The URL (case-insensitive) already exists
Search indexing is not configured on the server (Algolia credentials are missing).	ALGOLIA_APP_ID or ALGOLIA_ADMIN_API_KEY is unset
<host>/<path> was not found or is private; only public repositories are supported on <host>.	Non-GitHub host: the repository is missing or needs credentials
Can't index <host>/<path>: <host> resolves to a private or loopback address. or ...: <host> could not be resolved.	Non-GitHub host: the name points at a non-public address, or doesn’t resolve
GitHub repository octo/demo was not found. Check the URL; if it's private, the server's GITHUB_TOKEN needs access.	github.com: missing, or private without token access
GitHub denied access to octo/demo.	github.com answered 403
Branch 'x' does not exist in octo/demo.	The branch is missing
octo/demo is empty; there is nothing to index.	No commits
GitHub's API rate limit was reached; try again in about 2 minutes.	GitHub is rate limiting the server’s token
GitHub rejected the server's GITHUB_TOKEN; check the token configured on the CodeSage server.	The token is invalid
Could not reach GitHub; try again. or Could not reach <host>; try again.	The host or git failed in another way
GitHub did not respond in time; try again. or <host> did not respond in time; try again.	The 10-second check timed out

Example prompt

Add https://gitlab.com/group/project to CodeSage, on the dev branch, then tell me when it is ready.

Resources

The bridge also answers resources/list and resources/read. Reading needs the API key, because no public endpoint exposes the file list.

URI	MIME type	Listed by resources/list	Content
codesage://repositories	application/json	Yes (name Repositories)	The same object as list_repositories
codesage://repositories/{id}/files	application/json	No	The files of one repository

{id} is the numeric ID, 1 to 19 digits. Names are not accepted here. The files resource is readable even when the repository is not completed.

Real output for codesage://repositories/2/files, with files trimmed to the first two entries:

codesage://repositories/2/files

{
  "repository": "pmndrs/zustand",
  "total_files": 120,
  "truncated": false,
  "files": [
    {
      "file_path": ".github/DISCUSSION_TEMPLATE/bug-report.yml",
      "language": "yaml",
      "line_count": 17
    },
    {
      "file_path": ".github/FUNDING.yml",
      "language": "yaml",
      "line_count": 16
    }
  ]
}

Field	Meaning
total_files	Every stored file for the repository
truncated	true when total_files is above the 5,000-file cap, so only the first 5,000 by path are listed
files[]	file_path, language and line_count, ordered by path

Resource errors

Resource reads fail as MCP protocol errors, not tool errors. The backend answers 404 and the bridge words it as Resource or repository not found: <detail>.

Detail	Cause
Unknown resource '<uri>'	The URI is neither of the two forms above, or the ID has more than 19 digits
Repository 99 not found	No repository has that ID. IDs at or above 2^63 are treated the same way

Related pages

• MCP HTTP endpoints for the request and response envelope.
• Search for how ranking and filters work.
• Troubleshooting for symptoms and fixes.